feat(core): implement Pattern-Key taxonomy and recurrence deduplication engine

[harshsinghmp]

Classification

actionable-feature

Context & Rationale

Currently, repeated agent errors, tool failures, and user corrections can lead to duplicate memory records. A structured two-level key taxonomy (area.symptom) allows deterministic grouping, deduping, and recurrence tracking.

Key Capabilities & Scope

1. Two-Level Taxonomy (area.symptom):
   • Standardize areas: api, auth, build, config, deps, fs, net, runtime, shell, vcs, simplify, harden.
   • Symptoms must be generic (no hostnames, file paths, or versions): e.g. deps.module-not-found, runtime.type-error.
2. Schema & Deduplication Engine:
   • Add columns to SQLite memory store: pattern_key TEXT, recurrence_count INTEGER DEFAULT 1, first_seen TEXT, last_seen TEXT.
   • Implement "fold-don't-duplicate" logic: on new capture, search by pattern_key; if found, increment count and refresh timestamp.

Verification Criteria

• SQLite schema migration includes pattern_key and recurrence fields.
• Capturing identical error keys increments recurrence_count rather than inserting duplicate records.
• Passes test suite (bun test test/).