fix: rotate worker sessions after each LLM call and add recall error handling

[BYK]

Bug 1: "Multiple reasoning_opaque values received in a single response"

Worker sessions for distillation and curation were reused across multiple LLM calls via ensureWorkerSession(). Each call's assistant response (with reasoning/thinking parts) accumulated in the session history. When a subsequent call sent this history back, the provider rejected it with "Multiple reasoning_opaque values".

Fix: Delete the parent→worker mapping from the workerSessions Map immediately after reading the response in all 4 functions (distillSegment, metaDistill, curator.run, curator.consolidate). Each LLM call now gets a fresh session. Worker session IDs remain in the workerSessionIDs Set so shouldSkip() still recognizes them.

Bug 2: Recall returns nothing after an error

The recall tool's execute() in src/reflect.ts had no try/catch. If any of the three search calls threw, the entire tool execution failed and returned nothing.

Fix: Wrap each search call (temporal.search, searchDistillations, ltm.search) in independent try/catch blocks so partial results are returned even if one source fails. Errors are logged via log.error().

Changes

• src/distillation.ts: Add workerSessions.delete() after session.messages() in distillSegment() and metaDistill()
• src/curator.ts: Same pattern in run() and consolidate()
• src/reflect.ts: Add import * as log, wrap each search in try/catch with log.error()

All 194 existing tests pass.