Fix ESLint v8 crash and fatal error handling in verify_task_is_complete

[hiroshinishio]

Summary

• ESLint v8 crash fix: Read ESLint major version from package.json and only use --no-warn-ignored flag for v9+. Previously this flag crashed ESLint v8 repos with exit code 2, trapping the agent in an infinite verify loop until Lambda timeout (900s).
• ESLint fatal errors non-blocking: Exit code >= 2 (infrastructure issues like invalid CLI options, missing plugins) now returns success instead of reporting as lint_errors. These are not code issues the agent can fix.
• Extracted get_dependency_major_version: Reusable utility to read a dependency's major version from package.json (checks devDependencies then dependencies, strips semver prefixes).

Test plan

• 26 ESLint tests pass (including new v8/v9 --no-warn-ignored tests)
• 7 get_dependency_major_version tests pass (v8, v9, semver prefixes, missing package, invalid JSON)
• flake8 clean, pylint 10.00/10, pyright 0 errors

Social Media Post (GitAuto)

Found a bug that burned 900 seconds of Lambda time per PR on ESLint v8 repos. The --no-warn-ignored flag (v9 only) crashed ESLint v8, the agent saw "tests failed," retried, crashed again, looped until timeout. Fix: read the eslint version from package.json and skip the flag for v8.

Social Media Post (Wes)

A customer PR timed out at 900s. CloudWatch logs showed the agent running ESLint, crashing, retrying, crashing - four times in a row. We were passing --no-warn-ignored to ESLint 8, which doesn't support it. One version check in package.json broke the loop.