Add proxy_search_session_bodies for full-text body search in sessions

[yfe404]

Summary

• New proxy_search_session_bodies tool — decompresses and searches actual HTTP request/response bodies stored in persistent sessions, returning grep-like context snippets around matches
• Adds responseContentType to SessionIndexEntry for efficient content-type pre-filtering without loading full records from disk
• Updates proxy_query_session description to clarify it's metadata-only and point users to the new body search tool

Motivation

proxy_query_session text search only matches against index metadata (URL, hostname, path, exchangeId, matchedRuleId). Users searching for text inside response bodies (e.g., a price like "299,-" in HTML) get 0 results even though the content exists in the decompressed body. The in-memory proxy_search_traffic searches bodyPreview but only covers the live ring buffer (1000 entries, lost on restart).

Key design decisions

• Separate tool — proxy_query_session stays fast (index-only), body search is explicitly heavier
• Single file handle for bulk record reads (vs open/close per record)
• Preview fallback — works with both full and preview capture profiles; source field tells the agent which it got
• Binary safety — null-byte detection + known MIME prefix skip list
• Backward compatible — old sessions without responseContentType in index still work

Test plan

• 30 new unit tests covering core search, case sensitivity, preview fallback, all pre-filters, limits, edge cases, backward compat, snippet extraction
• tsc --noEmit passes
• npm run build succeeds
• Updated integration test tool count (76 → 77)