fix: live view re-entering files, plain-folder bind, and structured live-events need real-machine evidence

[pacphi]

Problem

The live session view has three open questions that only real-machine evidence can settle:

1. Re-entering files. When a transcript file re-enters the live window after a restart, it is read from its start, not from where the previous run stopped. Decide whether that's acceptable (and document it) or resume from a remembered offset.
2. Plain-folder bind. Binding live sessions to a plain (non-Git) folder has fixture tests but hasn't been observed on a real machine.
3. Structured live-events path. The structured live-events input has fixture evidence only, because nothing produces such a file yet. Either document it as experimental, or add a real producer.

What's needed

Observe items 1 and 2 on a real machine and fix what the evidence shows, then decide item 3.

Background

• Audit record docs/audits/2026-09-26-issues-237-238-239-verification-and-decisions.md, open items → Live.
• Moved out of the remediation program's Branch 9 on 2026-09-28, because it needs real-machine observation rather than code alone.

[pacphi]

PR #276 (eb963f503f806f80df7a8173d7c71314adbd7cfa) reconciles the three scoped questions:

1. Actual native-transcript replay on bounded re-entry preserves accepted/session/project totals, with failing-before/passing-after evidence. Offsets are bounded in memory; eviction or a new service process may replay. Persistent exactly-once delivery is not claimed.
2. A genuine Codex CLI 0.159.0 idle-fork process/transcript join was observed in a plain non-Git folder with the real survey. This involved no model turn, billing proof or browser claim.
3. Structured input is explicitly experimental without a verified producer.

The implementation is integrated in develop. Final human-review PR #285 at 04e3de3c336fce47ce85a6684b39608d916036cc carries the main-merge closure. This issue remains open until that human-approved merge.

Bounded integration evidence and scope reconciliation preserve the observations and limitations.