[FIX]: Evaluate probe verdicts over final traces

[spencrr]

Description

Changes behavioral probes to evaluate the verdict once over the completed trace. Probes no longer stop on a detected verdict by default; callers can provide stop_when when an online condition should define an earlier terminal trace.

This fixes sequence and prior-condition probes that cannot be resolved correctly from partial prefixes. Final verdict evidence is stored on Result.evaluation, online stop evidence remains on turns, zero-turn runs return ERROR, and summaries use terminal rationale or evidence. Terminal evaluation stays inside the active session.

Depends on #148. Because the branches live on a fork, this PR temporarily includes lower-layer diffs and targets main; those diffs disappear as dependencies merge.

Breaking changes

Behavioral change: Result.eval_results is normally empty for probes without stop_when; use Result.evaluation for verdict evidence. Adaptive drivers without an online stop receive no evaluator feedback and may run to max_turns.

Checklist

• pre-commit run --all-files passes
• Tests added for one-call verdict cadence, sequence and prior-condition probes, scopes, stopping, feedback, budgets, session ordering, summaries, and error results
• Documentation updated

Validation: 164 probe, driver, serialization, and report regressions pass; strict documentation build and static checks pass.

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.