Populate `tool_result_events` archive table from execution-graph + content sidecar

[willwashburn]

Context

PR #78 created the tool_result_events table in archive.sqlite but left it empty:

tool_result_events — table reserved (created, not populated) for the future content-sidecar bridge (#33) and execution-graph work.

The table is defined at packages/ledger/src/archive.ts (schema block, with the comment Reserved for the content-sidecar bridge (#33) and the execution-graph work). Columns: source, session_id, message_id, tool_use_id, call_index, event_index, status, content_length, content_hash, subagent_session_id, agent_id, event_source, ts.

The upstream data already exists or will exist:

• The content sidecar (Design: content sidecar store with retention and opt-out #33) landed in PR Content sidecar with retention and opt-out (#33) #43 / PR Add content capture for Codex and OpenCode parsers (#33 follow-up) #58 — ContentRecord rows for tool_result are written to ~/.relayburn/content/<sessionId>.jsonl for Claude, Codex, and OpenCode.
• The execution-graph substrate (PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77, refs Execution graph for passive readers: session relationships and tool-result event chronology #42) is in flight and adds ToolResultEventRecord to the canonical ledger via appendToolResultEvents / queryToolResultEvents. Schema columns line up with the archive table almost 1:1.

This issue is the thin connector: read the canonical ToolResultEventRecord lines (and optionally enrich content_length / content_hash from the content sidecar where the execution-graph record doesn't already carry them) and materialize them into the archive's tool_result_events rows during incremental build.

Proposal

1. In archive.ts, add a third pass to buildArchive() that handles LedgerLine kind tool_result_event (added by PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77). Parse via the new isToolResultEventLine guard and INSERT OR REPLACE INTO tool_result_events keyed on (source, session_id, message_id, tool_use_id, event_index).
2. The execution-graph record already carries contentLength and contentHash (per PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77's record shape). If a future record lacks them, fall back to reading the matching ContentRecord from the content sidecar.
3. Track build progress through the same archive_state.ledger_offset_bytes cursor — no separate cursor needed since both turn and tool-result-event lines live in the same ledger.jsonl.
4. On rebuild-from-zero, replay tool-result-event lines alongside turn / stamp / compaction lines.

Tests:

• Round-trip: append ToolResultEventRecord rows, run buildArchive(), assert they show up with correct keys, statuses, and content metadata.
• Idempotence: rebuilding twice yields the same row count.
• Mixed source: Claude + Codex + OpenCode events all materialize correctly.
• Schema-mismatch path: bumping ARCHIVE_VERSION clears and re-populates tool_result_events from the ledger.

Acceptance criteria

• tool_result_events is populated incrementally during burn archive build.
• All columns defined in the schema have non-null values where the upstream record provides them.
• burn archive status --json row counts include tool_result_events.
• Archive rebuild from a ledger with N tool-result events yields exactly N (deduped) rows.
• No regression to existing archive-build behavior for ledgers that pre-date PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77.

Out of scope

• Defining the ToolResultEventRecord shape (covered by PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77 / Execution graph for passive readers: session relationships and tool-result event chronology #42).
• Codex / OpenCode passive-reader population of those records (covered by PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77's deferred items).
• Consumer queries that read tool_result_events (waste detectors, subagent tree, etc.).

Refs

• Spun out of Derived analytics archive foundation: archive.sqlite + burn archive (#40) #78 (Derived analytics archive: materialize the ledger into a local queryable store #40 follow-up)
• Depends on PR Execution graph substrate: SessionRelationshipRecord + ToolResultEventRecord (#42) #77 / Execution graph for passive readers: session relationships and tool-result event chronology #42 landing the canonical ToolResultEventRecord writer path
• Related to Design: content sidecar store with retention and opt-out #33 (content sidecar — already merged via PR Content sidecar with retention and opt-out (#33) #43, Add content capture for Codex and OpenCode parsers (#33 follow-up) #58)