Fix stale parent directory sizes after subtree reindex

[wangruohui]

Summary

This PR fixes a stale-cache issue in duc when reindexing a subtree.

Previously, if a user indexed a, then later modified files under a/b and ran duc index a/b, the subtree report for a/b was refreshed, but the cached directory entry stored in a was not. As a result, commands like duc ui a or duc ls a could still show the old size for b.

Root Cause

duc stores directory listings as serialized per-directory buffers, and parent directories cache the size of each child directory entry. Reindexing a subtree updated:

• the subtree's own report
• the subtree's own directory buffer

but it did not propagate the new size back into ancestor directory buffers.

That meant readers opening a still saw the old cached size for b.

Changes

This PR adds parent-buffer propagation after writing a refreshed subtree report:

• update the immediate parent directory entry with the subtree's new exact size and devino
• compute the size delta between the old and new subtree values
• propagate that delta upward through ancestor directory buffers

This keeps cached parent views consistent after duc index <subtree> without forcing a full reindex of the ancestor tree.

Example

Before this change:

1. duc index a
2. modify files under a/b
3. duc index a/b
4. duc ui a

b could still show the old size inside a.

After this change, b is updated correctly in the parent view.

[Copilot]

Pull request overview

This PR fixes stale cached directory sizes when reindexing a subtree by propagating updated subtree sizes into ancestor serialized directory buffers, keeping parent views consistent without a full reindex.

Changes:

• Add helpers to split a path into parent/child and to update a child directory entry inside a parent directory buffer.
• After writing a refreshed subtree report, update the immediate parent entry (exact size + devino) and propagate size deltas up ancestor buffers.
• Hook the propagation into duc_index() after db_write_report().

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[l8gravely]

Thanks for sending in this patch, along with a test case! I need to look it over more but one nit I have is that I don't like how the args passed into fuctions are split one per-line now in some of the new functions you call. Not ideal in my feeling, it spreads the code out a bunch more.

I also see where you change some size_t back to int, which I think is wrong too, since we could potentially have large values there. But this will take some more work to look over.