fix(core): stable id tiebreaker in conversation list sort (kills the CI flake)

[matze4u]

The flake

handler-list.test.ts failed ~25-50% of CI node jobs — it forced a
re-run on PR #66 and PR #68. Root cause: BaseConversationStore.list
sorted newest-first by updatedAt alone:

.sort((a, b) => b.updatedAt.localeCompare(a.updatedAt))

Two conversations created/touched in the same millisecond compare equal,
so the resulting order is non-deterministic (['c1','c2'] vs
['c2','c1']).

Fix

Add id as a stable secondary key:

.sort((a, b) => b.updatedAt.localeCompare(a.updatedAt) || b.id.localeCompare(a.id))

Ids carry a monotonic counter (conv-<ts>-<n>-<rand>), so id-descending
≈ creation order — the tiebreak stays newest-first. The sort lives in the
shared BaseConversationStore, so both InMemoryConversationStore and
FileConversationStore are covered.

Verification

• New contract test freezes the clock so both conversations share
  updatedAt, then asserts deterministic order (id-2 before id-1) —
  runs for both stores.
• The previously-flaky handler-list.test.ts now passes 10/10 in a
  loop (was intermittently failing).
• Full core suite: 1157 passed / 1 skipped.