feat: persistent context eviction — context_evict events + eid identity (by Wren)

[conoremclaughlin]

Why

Implements Layer 2 of spec:sb-context-eviction (v2 — consolidated tonight with Conor). PR #242 gave SBs list_context/evict_context, but evictions were in-memory only: hydration replays raw transcript events, so evicted entries resurrected on reattach — the SB's choices didn't stick. This is the persistence foundation everything else (interactive UI, /evict, retention policies) builds on.

Storage decision (analyzed in the spec): event-sourced exclusion in the JSONL — append-only, local-first, same proven machinery as compaction events. Cloud mirror (Supabase projection) explicitly deferred.

What

• eid identity: every appendTranscript event gets a monotonic file-relative id; hydration seeds the counter from the file's max
• context_evict events: evict_context results include evictRefs (eid + content hash per removed entry); the runtime persists them with actor/reason
• In-stream replay: hydration applies evictions at their position in the file — an eviction only affects entries before it, so identical content appended later survives (ordering semantics for free, regression-tested)
• Ref matching: eid when present (precise); content-hash fallback for legacy/live entries (identical duplicates evict together — documented semantics)
• Trims persist too: /trim and the compaction hard-trim fallback write context_evict (actor: system) — closes the pre-existing "trims resurrect" gap
• Evicted ≠ erased: skipped entries collect into a side list; Ctrl+O inspector gains an "Evicted from Context" section with attribution
• Safety: evict events never touch entries that predate hydration (bootstrap); compaction keptEntries carry eids so kept-tail entries remain individually evictable

Verification

• 15 hydration tests (6 new eviction scenarios), 52/52 across hydration+ledger+tools, full CLI suite 788 passed (1 known gemini WIP failure)
• Live E2E: Myra ran list_context → evict_context source:old-heartbeat (5 stale entries, 430 tok freed) → reattach: "Old-heartbeat entries: None present… they have not returned. Launch codename: BLUEBIRD — still intact."

🤖 Generated with Claude Code

[conoremclaughlin]

Changes requested — two issues before this should merge:

1. packages/cli/src/repl/context-ledger.ts still contains a literal NUL byte in entryRefHash (${role}<NUL>${content}). chat.ts is clean, but this file is now treated as binary by Git (git diff --numstat origin/main...HEAD -- packages/cli/src/repl/context-ledger.ts prints -\t-). Please replace the separator with a textual escape / buffer construction so the checked-in TS source remains plain text, e.g. .update(role).update('\0').update(content) or equivalent.

2. The hydration ledger is correct for eid-specific duplicate evictions, but tailPreview is not: the preview-removal fallback applies removedKeys to every preview row, even when the row has a non-matching eid. Repro I ran: transcript {eid:1,user:'duplicate'}, {eid:2,user:'duplicate'}, then context_evict refs:[{eid:1}] leaves ledger entry eid 2 and messageCount=1, but tailPreview=[]. That means a surviving in-context turn disappears from the reattach replay. Please make preview filtering prefer eid when present and only use role/content fallback for legacy/no-eid rows, plus add a regression asserting tailPreview keeps the non-evicted duplicate.

Verified while reviewing:

• git diff --check origin/main...HEAD clean
• NUL scan: only packages/cli/src/repl/context-ledger.ts has a NUL byte; chat.ts has none
• ./node_modules/.bin/vitest run packages/cli/src/commands/chat-hydration.test.ts packages/cli/src/repl/context-ledger.test.ts packages/cli/src/repl/context-tools.test.ts → 52/52 passed
• yarn workspace @inklabs/cli type-check passed

[conoremclaughlin]

LGTM. Re-reviewed at 4ed81180; both prior blockers are fixed.

Confirmed:

• context-ledger.ts is text again: NUL scan clean across all five touched files, git diff --numstat origin/main...HEAD -- packages/cli/src/repl/context-ledger.ts reports normal line counts (41\t1) instead of binary -\t-, and git grep -I entryRefHash finds the file.
• Preview filtering is now eid-preferred: eided preview rows only match removed eids, and role/content fallback is limited to eid-less legacy removals.
• The exact duplicate-eid repro now leaves ledger eid 2, tailPreview eid 2, and messageCount=1.
• New regression test covers that case.

Verification run:

• git diff --check origin/main...HEAD
• NUL scan of the five touched files
• ./node_modules/.bin/vitest run packages/cli/src/commands/chat-hydration.test.ts packages/cli/src/repl/context-ledger.test.ts packages/cli/src/repl/context-tools.test.ts → 53/53 passed
• yarn workspace @inklabs/cli type-check

Prior inline thread resolved.