fix(parser): detect opencode invalid tool calls as failures

[ajinkyajacob]

Fixes #1254

Problem

Opencode stores failed tool calls with tool: "invalid" and state.status: "completed" — the parser never emitted a ResultEvent, so ComputeToolHealth had no error signal to detect. All opencode sessions showed tool_failure_signal_count: 0 despite confirmed invalid tool calls.

Fix

In extractOpenCodeToolCall(), when d.ToolName == "invalid", attach a ResultEvent{Status: "errored"} so the signal pipeline (extractToolCallRows → signals.IsFailure → ComputeToolHealth) counts it as a failure.

Test

Added TestParseOpenCodeDB_InvalidToolCall that seeds a session with an invalid tool part and asserts ResultEvents[0].Status == "errored".

All existing opencode tests continue to pass.

[roborev-ci]

roborev: Combined Review (70581d9)

The change is sound overall, but it needs a data-version bump to apply correctly to previously synced sessions.

Medium

• internal/db/db.go:324 — The parser now persists additional result events, but dataVersion remains 71. Previously synced, unchanged OpenCode sessions will be skipped, so their historical invalid calls will not be recognized as failures.
  • Fix: Bump and document the data version, then update the corresponding version assertion to trigger a non-destructive resync.

---

Reviewers: 2 done | Synthesis: codex, 7s | Total: 2m24s

[roborev-ci]

roborev: Combined Review (993a978)

Code review passed: no Medium, High, or Critical findings.

---

Reviewers: 2 done | Synthesis: codex, 6s | Total: 2m39s