GenerateTelosSummary: sample TELOS.md text parses as real content into the session-imported summary

[vichong]

When the legacy per-topic TELOS files are archived per the 2026-05-01 migration, TELOS.md becomes authoritative — and the shipped sample text ("(sample) The one-sentence reason you do what you do") parses cleanly into PRINCIPAL_TELOS.md, which is @-imported into every session. Sandbox-confirmed on v7.x: sample missions rendered as the principal's real missions.

Two related defects in the same path:

1. readTelosFile() checks the legacy per-topic path first and returns immediately — while those files exist, TELOS.md is never read at all, yet the output stamps derived_from: TELOS.md.
2. Consolidation is lossy. parseGoals() buckets on ## Active/## Deferred/## Completed H2 headings, but loadTelosSections() also splits TELOS.md on H2, so the buckets can't survive nesting under a ## Goals section. Sandbox: deferred and completed goals vanished entirely.

Suggested fix: refuse to parse entries containing the (sample) marker; extend the existing zero-item guard (~line 525) to also trip on sample-only content; document that per-topic files win while present.

[danielmiessler]

Partial fix landed; leaving this open for the rest. Done: the (sample) guard — both parse paths (ID-form and prose) now refuse shipped-template placeholder text, so sample missions can no longer render as the principal's real content (credit comment cites you). Still tracked from your report: the legacy-first readTelosFile() ordering vs the derived_from stamp, and the H2 bucket-nesting loss for ## Active/## Deferred/## Completed under ## Goals — both real, both need a more careful pass than a triage batch should carry. The sample guard ships in the next release.