feat(artifacts): normalize transcript sidecar contract

[christso]

Summary

Bead: av-vwa.16.8

• Adds the provider-neutral agentv.transcript.v1 JSONL row contract and derives eval artifact transcripts from the canonical agentv.trace.v1 envelope/projection path.
• Writes per-result outputs/trace.json and outputs/transcript.jsonl, with index rows pointing at the per-result transcript_path; removes the run-root aggregate transcript writer.
• Keeps provider-native evidence opaque under nested metadata/tool payload fields, preserves legacy transcript replay/import compatibility, and documents trace as the full-fidelity sidecar.
• Covers Codex-style artifact writing, Harbor/compact lifecycle evidence, Pi-style tool-call/tool-result pairing, replay fixtures, and dashboard transcript fixture discovery.

Verification

• bun --filter @agentv/core build - passed
• bun test packages/core/test/evaluation/trace-envelope.test.ts packages/core/test/import/transcript-provider.test.ts apps/cli/test/commands/eval/artifact-writer.test.ts apps/dashboard/src/components/transcript-timeline.test.tsx - passed, 65 pass / 0 fail
• bun run lint - passed
• bun run typecheck - passed
• rm -rf /tmp/agentv-transcript-sidecar-green && bun apps/cli/src/cli.ts eval examples/showcase/trace-evaluation/evals/coding-agent-replay.eval.yaml --target replay_coding_agent --output /tmp/agentv-transcript-sidecar-green && <artifact inspection> - passed; confirmed no run-root transcript.jsonl, per-result outputs/transcript.jsonl, per-result outputs/trace.json, agentv.trace.v1, and agentv.transcript.v1

Post-Deploy Monitoring & Validation

• No runtime service deployment is required. Validation should focus on CI plus inspecting a produced eval bundle for per-result outputs/trace.json and outputs/transcript.jsonl paths.
• Watch for downstream consumers still looking for run-root transcript.jsonl or outputs/execution-trace.json; those surfaces were same-week artifact contracts and have been renamed/removed here.

Residual Risks

• Live provider verification was not run; coverage is fixture-based plus replay smoke eval.
• Existing imported transcript JSONL without v1 fields remains replayable, but eval artifact consumers should move to index row transcript_path rather than root transcript discovery.