feat: add TTL-based file-backed caching for GitHubCLIProvider

[Achiever199]

Summary

Closes #35

Adds an optional caching layer to GitHubCLIProvider to avoid
redundant gh CLI calls within and across sessions.

Changes

New file: src/oss_dev/providers/cache.py

• CacheEntry — stores value + expiry timestamp
• ResponseCache — two-layer cache (memory + disk)
  • Memory layer avoids repeated disk reads in the same process
  • Disk layer (JSON files) survives process restarts
  • TTL-based expiry on all entries
  • make_key() uses SHA-256 hashing for safe filenames

Modified: src/oss_dev/providers/github/client.py

• Added _run_gh_cached() helper for read-only calls
• Cached methods: fetch_issue, list_issues,
  get_pr_status, get_pr_comments
• create_pr intentionally bypasses cache (mutating operation)
• Cache controlled via config.cache.enabled / ttl / dir

New file: tests/providers/test_cache.py

• 22 unit tests, all passing
• Covers: TTL expiry, disk persistence, disabled mode,
  invalidation, provider integration

[Hell1213]

Review: TTL-based caching for GitHubCLIProvider

Hi @Achiever199, great implementation! The cache is well-structured with 22 tests covering TTL, disk persistence, invalidation, and disabled mode. Tests pass (94/94).

Issues to fix (ruff F401/F841):

1. src/oss_dev/providers/cache.py:12 - import os is unused (the code uses pathlib instead)
2. tests/providers/test_cache.py:13 - MagicMock imported but unused (only patch is used)
3. tests/providers/test_cache.py:91 - Variable key assigned but never used
4. tests/providers/test_cache.py:193,210,227,240,256,272 - Each test imports GitHubCLIProvider locally, but _make_provider() already imports it. The per-test imports are redundant and trigger F401. Remove them.

Conflict notice: PR #54 by @Jiya3177 also modifies src/oss_dev/providers/github/client.py (adds rate limit awareness). Both change _run_gh and class internals. The author who merges second will need to rebase and reconcile.

Suggestion: Consider whether you want _run_gh_cached to also respect rate limits (from PR #54) once both are merged.