DB-backed sync, watermarks, and Pinecone preprocessors

[snowfox1003]

Summary

Integrate clang_github_tracker with the database: persist llvm/llvm-project issues, PRs, and commits during sync, use DB watermarks for GitHub API resume (replacing state.json), and drive Pinecone issue/PR selection from ClangGithubIssueItem.updated_at vs PineconeSyncStatus.final_sync_at (plus failed_ids retries). Raw JSON under workspace/raw/github_activity_tracker/ is unchanged.

Scope

• Models (no FKs): ClangGithubIssueItem (number, is_pull_request, github_* timestamps, created_at / updated_at); ClangGithubCommit (sha, github_committed_at, audit timestamps).
• Services: upsert_issue_item, upsert_commit, get_issue_item_watermark, get_commit_watermark, start_after_watermark (API cursors use Max(github_updated_at) / Max(github_committed_at)).
• Sync: sync_raw_only upserts after save_*_raw_source; single start_item for unified issues+PRs fetch.
• Date resolution: resolve_start_end_dates — DB-only; Boost-style --since / --until (aliases, invalid range → warning + clear bounds).
• CLI: run_clang_github_tracker aligned with Boost (--skip-*, --dry-run); removed --no-upload, --upload-only, Pinecone CLI overrides; logger-only output in package.
• Preprocessors: DB query updated_at__gt=final_sync_at (or all rows if final_sync_at is None), union failed_ids, then build_issue_document / build_pr_document from raw files. Do not change cppa_pinecone_sync or github_preprocess scan helpers.
• Backfill: backfill_clang_github_tracker — --from-csv (default path under workspace/clang_github_tracker/) xor --from-raw.

Docs

• docs/Workspace.md, Schema.md, Pinecone_preprocess_guideline.md, service_api/README.md, service_api/clang_github_tracker.md
• Align Deployment.md / operations/github.md if they still mention state.json or removed clang flags.

Notes

• github_updated_at: API fetch watermarks only.
• updated_at: Pinecone incrementality only.

[snowfox1003]

I create a pr - #137