feat: add incremental indexing during save operations

[asvarnon]

Summary

Add incremental indexing at save time so new entries update term/frequency state without full corpus recomputation.

Motivation

Save-time processing must stay fast (<200ms typical) while keeping analysis state query-ready. Incremental updates avoid requiring full cf analyze runs between saves and queries.

Scope

In scope:

• Hook incremental analysis into save_snapshot() for cf save and cf pre-compact
• Tokenize only the new entry
• Update SF/frequency state for affected terms only
• Keep incremental state consistent with batch recomputation semantics
• Likely files: crates/core/src/engine.rs, crates/analysis/src/* incremental methods

Out of scope:

• Full corpus recompute path (cf analyze)
• Save-time extraction/classification beyond agreed incremental subset

Acceptance Criteria

• Each save triggers incremental tokenization for the new entry only
• SF counts update correctly for terms in the new entry
• Typical save-path latency impact remains under 200ms
• Benchmark test compares incremental update vs full recompute
• Consistency test shows incremental state matches full recompute results

Dependencies

• feat: add analysis crate foundation and term extraction #46
• feat: implement session frequency and recurrence scoring #47

Complexity

Medium — correctness and performance constraints must both hold under incremental updates.

[asvarnon]

Status update (2026-04-03): All dependencies merged — #46 (PR #59), #47 (PR #64). This issue is now unblocked and ready to implement.