docs(use-arc): add Common Pitfalls section with empirical gotchas

[Ccheh]

Adds a new "Common Pitfalls" section to the `use-arc` skill capturing four real gotchas hit while building production-grade contracts and clients against Arc Testnet:

1. Native USDC vs ERC-20 USDC decimals — the most common bug. Agents default to `parseUnits(amount, 6)` because USDC is 6 decimals on every other chain. For native gas on Arc it must be 18. Includes wrong/right code examples + symptom description.
2. `eth_getLogs` is capped at 10,000 blocks per call — RPC error `-32614: eth_getLogs is limited to a 10,000 range`. Includes a chunked-backward enumeration pattern.
3. `waitForTransactionReceipt` frequently times out — Arc Testnet's RPC drops receipt-query connections for 30-60s windows. Treating timeouts as failures causes false-negative reports. Includes a poll-with-tolerance pattern + the "pre-compute deterministic IDs" trick.
4. Avoid `--private-key` for long-running scripts — the keystore-via-`cast wallet import` pattern. The current skill mentions this rule in passing in the Foundry block; this version adds the concrete `cast wallet import deployer --interactive` + `forge script --account` flow.

Why this matters

The `use-arc` skill is the single piece of documentation an AI agent reads when generating Arc code. Currently it correctly states the rule "use 18 decimals for native gas amounts and 6 decimals for ERC-20 USDC amounts" — but doesn't give code examples showing what the wrong pattern looks like, or what symptoms it produces. Agents that read the rule still consistently produce `parseUnits(amount, 6)` because the prior "USDC is 6 decimals everywhere" is reinforced by ~8 chains of correct-elsewhere training data. Showing the wrong code → right code transition makes the skill behaviorally effective, not just descriptively correct.

The other three pitfalls (`eth_getLogs` cap, receipt timeout pattern, keystore over CLI flag) are operational realities of Arc Testnet that aren't documented anywhere agent-readable but bite every team building real applications.

Source of these pitfalls

Encountered while building Plinth — a capital layer for AI trading agents on Arc — over the Agora Agents Hackathon (May 2026). All four patterns were debugged against the live testnet RPC; the code examples come from the working production codebase.

Diff size

+4,382 chars added in a single new section between "Implementation Patterns" and "Next Steps". No existing content modified.

Happy to iterate on placement, depth, or wording. If maintainers prefer this as a separate `arc-pitfalls` skill rather than an inline section, I'm equally happy to split it.