feat: Benchmark Dataset Catalog (M120)

[hlin99]

Summary

Add a SQLite-backed local catalog for indexing and searching benchmark files.

Changes

• DatasetCatalog class in catalog.py with SQLite storage
• CatalogEntry, CatalogQuery, CatalogReport Pydantic models
• SHA-256 file hash for duplicate detection
• Metadata extraction: GPU type, model name, P:D ratio, QPS, request count, instances
• Search/filter by GPU type, QPS range, P:D ratio, model name, instance count range
• CLI catalog subcommand: add, list, search, show, remove
• Programmatic manage_catalog() API
• 23 new tests
• Updated docs/iterations/current.md

Closes #263

[hlin99-Review-Bot]

✅ Approved by hlin99-Review-Bot

Idea Value: Strong. A SQLite-backed benchmark catalog is a natural next step — makes benchmark data discoverable and queryable instead of scattered files. Good alignment with the project's benchmarking infrastructure.

Code Quality:

• Clean SQLite schema with proper indexes for common query patterns
• SHA-256 duplicate detection ✓
• Comprehensive search with multiple filter dimensions
• CLI well-structured with Rich table output + JSON format option
• 23 tests covering add/remove/search/duplicates/edge cases ✓
• docs/iterations/current.md updated ✓
• CI all green (lint + tests on 3.10/3.11/3.12) ✓

LGTM 🚀

[hlin99-Review-BotX]

✅ Approved by hlin99-Review-BotX

Idea Value: Good fit. A local SQLite catalog for benchmark files is a natural extension — makes discovery and filtering straightforward without external dependencies.

Code Quality:

• Clean SQLite schema with proper indexes for gpu_type, pd_ratio, qps, date ✓
• SHA-256 duplicate detection ✓
• Pydantic models well-structured ✓
• Search supports multiple filter dimensions with proper parameterized queries ✓
• CLI with Rich table + JSON output ✓
• 23 tests ✓
• docs/iterations/current.md updated ✓
• CI all green (lint + tests 3.10/3.11/3.12) ✓

LGTM 🚀