chore: untrack hoot.exe + hoot-test, expand .gitignore

[oth-body]

Summary

Issue #50 was closed in July 2026 with the title "Remove binary from git and add .gitignore" — but the binaries were never actually untracked. This PR does what #50 claimed to do.

What this fixes

26 MB of build artifacts were tracked in master:

• hoot.exe — 13,054,464 bytes (Windows build of hoot)
• hoot-test — 13,012,992 bytes (Linux binary built by hoot_test.go's TestMain for black-box tests)

That's 26 MB bloating every clone, every git fetch, every CI checkout, and every PR diff stat — for files that have no business being in source control.

How

1. git rm --cached hoot.exe hoot-test — files stay on disk for anyone with a pre-existing clone, but stop syncing future changes.
2. Rewrote .gitignore (was just .aider*) to cover:
   • All common Go build outputs: hoot, hoot.exe, hoot-test, hoot-test.exe, hoot-cli, plus *.exe/*.so/*.dylib for renamed/relocated binaries
   • Test artifacts: *.test, *.out, coverage.txt, coverage.html, *.coverprofile
   • Go workspace files: go.work, go.work.sum
   • Standard IDE scratch files: .aider*, .idea/, .vscode/, *.swp, *.swo, .DS_Store
   • Dependency caches: vendor/
   • Local config overrides: .env, .env.local, *.local

No history rewrite

The binaries are still in past commits for anyone who needs to bisect through them. This is the minimum surgical change — stop the bleeding from here. If you want a clean history later, git filter-repo (or BFG) can purge the blob references; that's a separate, more invasive change that requires coordination with all clones.

Conflict check

• fix(hoot): dedupe defaultRelays; surface NWC payment uncertainty #61 (silent-failures-hoot): zero overlap — fix(hoot): dedupe defaultRelays; surface NWC payment uncertainty #61 modified defaultRelays, generateProfileID, payInvoiceNWC. This PR touches .gitignore and removes two tracked files.
• fix(hoot): publishNote counts successes; error path no longer skips defers #62 (publish-error-surfacing): zero overlap.
• fix(tipping): validate NWC URI before saving; add HTTP timeouts on LNURL #63 (tipping-path-hardening): zero overlap.

Related but not in this PR

• The prior refactor sprint (refactor: extract more CLI commands into error-returning functions #37, refactor: eliminate all log.Fatal calls, standardize CLI error handling #38) and this audit's fix(hoot): publishNote counts successes; error path no longer skips defers #62 fixed the log.Fatalf defer-skipping bug on the publish path. There are still 12 remaining log.Fatalf calls in main() that bypass defer eventCache.Close() on the failure path — same shape as fix(hoot): publishNote counts successes; error path no longer skips defers #62's fix, mechanical follow-up. Worth its own PR so this one stays focused.
• The new hoot_test.go test (TestNWCParsing from fix(tipping): validate NWC URI before saving; add HTTP timeouts on LNURL #63) requires Go to run. The repo has no CI workflow yet — adding .github/workflows/ci.yml running go test -race ./... would be a clean next step.