fix(sessions): preserve every distinct history blocker - #4715

Root cause

pass_reason_code retained only the first matching failure for each provider with BTreeMap::entry(...).or_insert(...). A provider reporting an invalid observation followed by projection_store_corrupt therefore lost the corruption reason before the worker logged the outcome or published its typed serving status. Source order could change the operator-visible diagnosis while the actual failures were identical.

Change

Collect distinct (provider, reason) pairs in canonical order rather than one reason per provider. Preserve raw codes for single-provider passes and existing provider-qualified codes for multi-provider passes. Duplicate failures do not repeat the same reason. Nonretryable failures still take precedence over retryable siblings, and healthy-provider progress remains intact.

This is the existing history-outcome/status path; no new event store, timer, retry policy, wire field, authorization rule, or recovery mechanism is introduced.

Regression coverage

Added three behavioral tests through the real outcome classifier:

• Multiple terminal failures from one provider retain corruption in either source order, deduplicate repeated failures, reach the actual SessionTemporalRefreshWakeState serving-status handle, and clear on a subsequent Complete outcome.
• Multiple retryable failures from one provider retain every distinct reason in stable order without changing retry classification.
• Mixed providers preserve qualified terminal reasons and committed progress while excluding retryable siblings from the blocked diagnosis.

Existing single-reason, mixed-provider, progress, and retry tests remain unchanged.

Verification

Reviewed the project/profile ingestors, outcome classifier, worker consumer, and typed status mapping. The complete GitHub-generated commit diff changes only reason aggregation, its documentation, test imports, and the three added regressions.

No local compilation, rustfmt, Clippy, or test execution is claimed. This environment has no supplied checkout, Bazel, rustc, or rustfmt, and direct source downloads fail DNS. No Cargo commands were run. Hosted Bazel CI must validate //crates/tracedecay-session-runtime:unit_test and the affected session/status journeys before merge.

Refs #4569 (corrupt-store visibility). This fixes information loss in the existing status path; it does not implement relational-store rebuild, prove a production corruption recovery, or complete Doctor's remaining recovery guidance. Keep the issue open.