[DanielNetzer]

Implements issue #414.

Molecule: mol-414-994b228e1d39479a56cd6d4e
Formula: forgecell-github-issue
Base: e0f33dfe1fe5ca921a28730611f92e1e30d5e3fa
Delivery: 6641d9251a2475cecfd9419eaa7e9b43eecfe0cf

Awaiting exact-commit CI and human review. Harness-reported checks are not independent CI evidence. No merge, deployment or issue closure is authorized.

[DanielNetzer]

Review notes (LOW, not blocking):

• --rebind and --model proposals on a Formula that declares harnesses.analysis still record the default harness's workflow in their saved evidence (saveWithEvidence now keeps the caller's value, and those callers pass the default's). Same as before this PR, so not a regression; doctor and the init summary read the analysis role correctly.
• TestInitReportsCursorBeforeApprovalWithoutFallback (init_test.go) pinned the old Cursor bootstrap and was updated in scope; its protected-view failure was accepted with --accept-protected-failures, the first live use of the --accept-protected-failures refuses when a costlier regression check comes first in the plan #406 fix in v0.4.7.