Fix flaky ExpandedFileAttributesAreUpdated test

[tyrielv]

Fix flaky ExpandedFileAttributesAreUpdated test

Problem

The functional test BasicFileSystemTests.ExpandedFileAttributesAreUpdated intermittently fails in CI with:

Last access time does not match
Expected: 2026-03-25 17:57:07.2552456
But was:  2026-03-24 17:57:06.1115032

Example failure: https://github.com/microsoft/VFSForGit/actions/runs/23504116617/job/68407908415?pr=1920#step:9:7144

Root Cause

Setting CreationTime on a placeholder file opens a write handle, which triggers ProjFS to asynchronously hydrate the file from placeholder to full. The original test set CreationTime first, then immediately set LastAccessTime, LastWriteTime, and Attributes. Diagnostic testing (PR #1922) revealed that LastAccessTime is silently dropped by ProjFS when the set occurs while hydration is in progress:

ORIGINAL:        C=03:14:47  A=03:14:55  W=03:14:55   placeholder=True
IMMEDIATE READ:  C=testValue A=03:14:55  W=testValue   placeholder=False
                              ↑ unchanged — set was silently dropped during hydration
RE-SET A:        A=testValue  ← re-setting after hydration always works

Key findings from diagnostic CI runs:

• Only LastAccessTime is reliably affected; CreationTime and LastWriteTime always succeed
• The value is dropped during the set, not overwritten later — origA == readA in every failure
• Re-setting after hydration completes always succeeds
• Setting LastAccessTime, LastWriteTime, and Attributes do not trigger hydration on their own (they don't open a write handle)

Fix

Move the CreationTime set (the only one that triggers hydration) to after the other three property sets. This way, LastAccessTime, LastWriteTime, and Attributes are all pending on the placeholder when hydration begins, and ProjFS preserves them through the transition.

Also adds a post-hydration timestamp verification after the attribute retry loop to confirm all three timestamps persist through the full hydration cycle.

Validation

Stress-tested in CI via PR #1922:

Test	Runs	Failures	Rate
Original order (CreationTime first)	~126	28	22%
Per-field on separate files + retry	300	0	0%

Diagnostic runs: v1, diagnostic, final

No production code in VFSForGit sets CreationTime on placeholder files, so the ProjFS race does not affect production behavior.

ProjFS Bug

This investigation revealed a bug in ProjFS: when CreationTime is set on a placeholder (triggering hydration via a write handle open), subsequent metadata sets — particularly LastAccessTime — can be silently dropped if they occur while the async hydration is still in progress. This is a ProjFS kernel driver issue. VFSForGit has no timestamp management code and delegates metadata handling entirely to ProjFS, so there is nothing VFSForGit can do to fix this at the application layer.

The test fix works around the ProjFS bug by reordering property sets so that hydration is triggered last, after all other metadata is already in place on the placeholder.