Compact charter review context reports success with unresolved selected directive stems on main

## Problem

On unchanged main at ab4636831945d20ac06d7e9b61d832ffd9a0359b, compact review-context loading reports success and exits zero while governance resolution fails: charter-selected directive filename stems are treated as unavailable. The payload has references_count=0. Initial bootstrap output instead contains 235 references, masking the failure until subsequent calls.

## Reproduction

From a fresh Spec Kitty checkout with its configured charter and warmed environment:

```sh
SPEC_KITTY_ENABLE_SAAS_SYNC=0 .venv/bin/spec-kitty charter context --action review --json --no-mark-loaded
```

After the initial bootstrap has been marked loaded, inspect compact payload diagnostics for `governance unresolved` and `Charter selected unavailable directive(s)`, including `001-architectural-integrity-standard`.

## Evidence and scope

Independently reproduced during the read-only debugger-debbie review of #4194, both at unchanged base ab4636831 and PR head a8f593c. This is not introduced by #4194. Profile CLI loading still succeeds; reading the initial bootstrap is not proof that subsequent live governance resolution succeeds.

Local evidence: squad-debugger-base-context.json and squad-debugger-context.json in /tmp/spec-kitty-20260910-073221-0BR8RJ. No context checks or activation validation were weakened in #4194.

## Expected

Resolve valid configured directive selections consistently in compact and bootstrap modes, and make unresolved governance observable as a failed result rather than an apparently successful empty context.

Run the provided `spec-kitty charter context` command on unchanged main, then inspect compact payload diagnostics for `governance unresolved` and `Charter selected unavailable directive(s)`, including `001-architectural-integrity-standard`. Compare the compact result after bootstrap loading with the initial 235-reference payload; done means valid configured directives resolve consistently and unresolved governance produces a failed result instead of an empty successful context.

Written by the indexing model from the issue text.