fix(ci): unbreak the Go test suite on develop

[tieubao]

What's this PR does ?

Fixes the CI test suite (ci-test.yml / make test), red on develop since the Alchemy pricing revamp. Test-only, zero production code changed.

Which UI commands affect by this change ?

None (test-only change).

DoD

• Is it public API in documentation? Any docs update required ?, No, test-only.
• Does it change the response data shape ?, No.
• Is it require to update the mock data in preview ?, No.
• Include the sample call / example in this PR, N/A, see Reproduce section below.

Summary

ci-test (make test) has been red on develop since the Alchemy pricing revamp (mochi-pricing-revamp sub-goal 05) retired Etherscan/SolScan/CoinGecko pricing for a tiered resolver (Alchemy by-symbol/by-contract -> DexScreener by-contract). The tests below never got their mocks/fixtures updated for the new call paths. This PR is test-only, zero production code changed.

• Test_dexscreener_Get (pkg/service/dexscreener): fixture pair address (0x5201...be4d) was never a real dexscreener pair (confirmed live: the API returns pairs:null for it), pre-existing bad fixture, unrelated to the revamp. Swapped for the Uniswap V3 WETH/USDC pool, a deep-liquidity, long-lived pair, matching this file's existing convention of live unmocked calls.
• TestEntity_SearchCoins / TestEntity_UpsertCustomToken (pkg/entities): repo.CoingeckoSupportedTokens.GetOne() never returns a nil token in production (pg.go always returns a non-nil pointer, even on gorm.ErrRecordNotFound); the CoinGecko-era mocks returned literal nil, nil-deref'ing at SearchCoins' token.ID check now that this path is actually reached. The new price_resolver.go tier also does a second GetOne call per priced token (to resolve its Alchemy ticker symbol), needed a new mock. Added an Alchemy service mock, and found + fixed a nil e.log in one test's Entity fixture that the new logTierCounts call nil-panics on.
• TestHandler_SearchCoins (pkg/handler/defi): same root cause at the handler/DB-integration level. Mocked svc.Alchemy the same way, and updated the testdata/search_coin/200-ok-multiple.json golden fixture: two coingecko ids sharing symbol "doge" (binance-peg-dogecoin / dogecoin) used to get different hardcoded prices under old CoinGecko per-id batch pricing (0.1 / 0.2); the new Alchemy-by-symbol tier can't distinguish them, so both now expect the same price. New expected values were captured from a real run against the seeded test DB, not guessed.
• TestHandler_GetCoin was already fixed (skip-with-comment, "re-sourced off CoinGecko in sub-goal 05; see TestBuildCoinResponseByID") by a prior commit already on develop; no change needed.

Scope note: the task brief named TestHandler_GetCoin as the 4th failing test, but it was already green (SKIP) on develop. CL-015's own backlog description names TestHandler_SearchCoins instead, which was genuinely still red, fixed here too so the branch actually earns its name.

Before / after

go test ./... (full suite, mise-pinned Go 1.24.2 to match go.mod's go 1.19 toolchain requirement, matching ci-test.yml's pinned toolchain approach):

	Before (develop @ 8eb11ba)	After (this branch)
pkg/entities	FAIL (panic: nil deref in TestEntity_UpsertCustomToken, aborts the whole package before TestEntity_SearchCoins even completes)	ok
pkg/service/dexscreener	FAIL (Test_dexscreener_Get: dead fixture, real API returns no pairs)	ok
pkg/handler/defi	FAIL (TestHandler_SearchCoins: price mismatch)	ok
Full suite	3 FAIL packages	0 FAIL packages (244 package results, all ok / no-test-files)

Confirmed via git stash before/after comparison on the same checkout (not just the 4 named tests in isolation).

Coverage (touched packages/functions; zero production lines changed)

Function	Before	After
pkg/entities (whole package, all tests)	build/panic, no coverage obtainable	10.0%
SearchCoins (defi.go)	n/a (panic)	56.3%
CreateCustomToken (tokens.go)	n/a (panic)	48.6%
enrichPriceQuery (price_resolver.go)	n/a (panic)	50.0%
symbolForCoingeckoID	n/a (panic)	80.0%
tierAlchemy	n/a (panic)	66.7%
GetTokenPricesByID	n/a (panic)	92.3%
resolvePriceOne	n/a (panic)	69.2%
pkg/service/dexscreener (package)	38.2%	41.2%
GetByChainAndPairAddress	(only via a failing test)	61.5%

Per-function coverage on SearchCoins/CreateCustomToken/enrichPriceQuery/tierAlchemy sits in the 48-67% range, below an 80% bar. These are large, multi-branch functions (skull-symbol override, default-ticker/native/is_popular tie-break branches, DexScreener-fallback path) where the uncovered branches are pre-existing and unrelated to what broke CI. Reviewer 1 (test-coverage lens) assessed this as acceptable for a CI-red bugfix ("scope here is explicitly making CI green, not raising coverage") and flagged the tierAlchemy DexScreener-fallback branch as a cheap follow-up, not a blocker for this PR.

Review

Three review passes ran against this diff before opening:

• kit:code-reviewer (test-coverage lens): PASS-WITH-NITS, no blocking findings. Confirmed mocks are correctly scoped (the gomock.Any() matchers are compensated by exact-match dispatch inside their DoAndReturn closures), the same-price fixture change is a legitimate reflection of the new architecture, and coverage levels are acceptable for this task's scope.
• kit:security-reviewer: PASS. No secrets, no PII, no weakened security assertions. The one live unmocked network call (dexscreener) targets a public well-known Uniswap pool, matching this file's pre-existing convention.
• kit:advisor (fable, critique mode): 4 findings, all addressed in the second commit:
  1. Branch didn't actually make CI green (TestHandler_SearchCoins still red), fixed, now 0 FAIL packages.
  2. "Architecturally indistinguishable" comment overclaimed; it's the current resolver's tier-ordering choice (by-symbol before by-contract), not an inherent limit, softened the comment and flagged it as worth an explicit accept-or-fix decision from whoever owns the pricing revamp (see below).
  3. Dead coinIds/coinPrices fixture fields left over from the pre-revamp mocks, deleted.
  4. A tests[0]-vs-loop-scoped-tt coupling in tokens_test.go's Alchemy mock, commented for the next person adding a test case.

Worth a follow-up decision (not blocking this PR)

Two coingecko ids that share one ticker symbol (ethereum/ethereum-wormhole, binance-peg-dogecoin/dogecoin) now resolve to the same Alchemy-sourced price, where the old CoinGecko-based pricing could tell them apart by id. This PR's tests document that as today's actual behavior rather than trying to fake differentiation. Whoever owns the pricing revamp should confirm whether that's an accepted tradeoff or worth a DexScreener-by-contract fallback for same-symbol tokens.

Not in scope

• Deleting the 3 dead Etherscan/BscScan/Infura config fields (CL-014/CL-010 residual cleanup).
• Raising SearchCoins/tierAlchemy/etc. coverage beyond what's needed to get CI green (see coverage table above).

Reproduce

export PATH="$HOME/.local/share/mise/installs/go/1.24.2/bin:$PATH"   # pin to go.mod's toolchain, avoid brew-go GOROOT mismatch
docker-compose up -d postgres_test redis chrome
sql-migrate up -env=test
PROJECT_PATH=$(pwd) go test ./...

This is a production service repo; this PR is not being merged by me. Waiting on human review.