test(strategy): add red tests for redo-inheritance edge cases

The redo half of the adversarial review of #2574 (split from 4a5df627d0; the squash half landed with the squash PR). Each test asserts the linkage a user would want.

Failing on the redo commit (8):

• Redo inherits on a coincidental match: identical lockfile, convergent deletion, a window that ordinary commits never close, undoing a rebase then taking an upstream file.
• Redo walk passes through merge commits into history already on main (reset --soft main after merging main).
• The editor flow condenses into the inherited trailer that lands after the stamp.
• Links still lost: amend fold, and unstage after reset --soft overwriting ORIG_HEAD.

Passing guards (4): unrelated commit after undo, stash after rebase, -m redo condenses into the stamp, declined redo refused from writing into its own earlier checkpoint.