Flip fi commit-message write default legacy → terse

[chris-peterson]

Context

git-fi records the branches merged into fi in fi's commit message and reads that list back on the next run. Two formats exist: the terse compact encoding ((feature-a, feature-b)@[shorthash]) and a legacy standard git-merge message.

git-fi reads both formats (BL-02, BL-03). During the migration rollout it writes the legacy format for every fi commit — bootstrap, empty, or existing — so downstream consumers that parse the fi commit message keep working. The write format is a single switch: DEFAULT_WRITE_FORMAT in src/merge.ts (BL-04).

This issue is the reminder to flip that switch to terse once downstream tooling handles it — targeted for ~September 1, 2026.

Proposed approach

1. Confirm downstream consumers of the fi commit message parse the terse format (or no longer parse the message).
2. In src/merge.ts, flip DEFAULT_WRITE_FORMAT from "legacy" to "terse".
3. Update the "currently legacy" framing:
   • SPEC.md — the BL-04 rollout note.
   • src/merge.ts — the DEFAULT_WRITE_FORMAT comment.
   • docs/merge-process.md — the "currently writes legacy" wording.

Reading logic does not change — both formats stay readable indefinitely.

Acceptance criteria

• New fi commits are written in the terse format.
• Both formats still read and round-trip; the suite passes (update the format tests that assert legacy-on-write).
• SPEC.md BL-04, the merge.ts comment, and docs/merge-process.md no longer describe legacy as the current default.

Considerations

• Gate on downstream readiness, not just the calendar — Sept 1 is a target, not a hard deadline.
• No data migration needed: only newly written messages change.