vfs/cache: Fix promoteToFast race with WriteCloser Close

promoteToFast read FileInfo.Size from Stat after the VFS lock was
released, while WriteCloser.Close updated Size on the same live
object. Concurrent Create/Open on overlapping keys tripped the race
detector.

Stat now returns a FileInfo snapshot, and promotion uses the
ReadAll length for the fast-tier Create size. Promotion stays
best-effort.

#21

vfs/cache/cache.go

@@ -186,7 +186,7 @@ func (tc *TieredCache) Capacity() int64 {
 func (tc *TieredCache) promoteToFast(key string, reader io.ReadCloser) {
 	defer func() { _ = reader.Close() }() // best-effort close; error secondary to promotion attempt (async best-effort path)
 
-	// Get file info from slow tier to determine size
+	// Size for the space/ReadAll guards comes from a Stat snapshot, not the live in-map FileInfo.
 	var size int64
 	if slow := tc.slow.Load(); slow != nil {
 		if vfs, ok := slow.(vfs.VFS); ok {
@@ -210,7 +210,7 @@ func (tc *TieredCache) promoteToFast(key string, reader io.ReadCloser) {
 	}
 
 	// Guard promotion ReadAll using already-fetched size (in addition to space check above)
-	if size > 0 && size > (1<<30) { // conservative 1GB hard limit on promotion reads (aligns with typical max_object_size)
+	if size > (1 << 30) { // conservative 1GB hard limit on promotion reads (aligns with typical max_object_size)
 		return
 	}
 	// Read the entire file content
@@ -218,6 +218,8 @@ func (tc *TieredCache) promoteToFast(key string, reader io.ReadCloser) {
 	if err != nil {
 		return // Skip promotion if read fails
 	}
+	// Create with the bytes we actually hold so we never reuse a live FileInfo.Size.
+	size = int64(len(content))
 
 	// Create the file in fast tier
 	if fast := tc.fast.Load(); fast != nil {