Detect and reject individual no-effect changes after cherry-pick

[mtsgrd]

Problem

When committing a mix of changes to a branch, some changes could be silently dropped — the commit succeeded, but the caller received no rejection feedback for the omitted changes.

Example: deleting a file that only exists in another stack's workspace tree. The deletion has no effect on this branch's tree, but was not reported as rejected.

Root cause

The post-cherry-pick check only compared entire trees:

if tree_with_changes == target_tree { reject all }

When at least one change was effective the trees differed, so individual no-ops slipped through undetected.

Fix

After the cherry-pick, check each change individually via an entries_match helper that compares both mode and object_id at a given path between the target and result trees:

for each change:
  if entries_match(target, result, change.path)
     && entries_match(target, result, change.previous_path):
    reject as NoEffectiveChanges

Comparing mode (not just object_id) catches mode-only changes like executable bit toggles. Checking previous_path catches renames where only the source removal was effective.

Test

A new two-stacks-with-extra-file fixture verifies that committing a file deletion from another stack alongside a valid modification:

• rejects the deletion as NoEffectiveChanges
• still produces a commit for the modification

[Copilot]

Pull request overview

This PR improves commit creation when applying worktree changes onto a different target tree (cherry-pick/merge-trees step), by detecting and rejecting individual change requests that become no-ops after the cherry-pick—rather than only rejecting when the overall resulting tree is unchanged.

Changes:

• Add post–cherry-pick per-path checks to flag individual DiffSpecs as NoEffectiveChanges when the target and result tree entries match.
• Add a new fixture with two stacks where one stack introduces an extra file.
• Add a regression test ensuring a deletion of a file that only exists in another stack is rejected while other valid changes still create a commit.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
crates/but-core/src/tree/mod.rs	Adds per-change post–cherry-pick no-op detection and an early-exit when all changes are rejected (without cherry-pick conflicts).
crates/but-workspace/tests/workspace/commit_engine/new_commit.rs	Adds regression test for per-change rejection of a no-op deletion after cherry-pick.
crates/but-workspace/tests/fixtures/scenario/two-stacks-with-extra-file.sh	Adds fixture scenario with two stacks where only one stack contains an extra file.

[Byron]

Thanks a lot!

I took a close look and think this is sound, despite this only being an armchair review.
Please see my one comment that might warrant a change, but it's more of a nit.

As a general note, these lower-level internal functions tend to not be well documented on as a function (along their parameters), and I think it's always worth fixing this while one is at it.
These days, I treat the documentation as a big part of specifying intent, with the code being replaceable.

[Byron]

I think it makes sense to put this into the previous if needs_cherry_pick branch, instead of opening a new one.