Recreate blobs directory after store schema migration

[rusik]

Problem

When LoggerStore is initialized against an existing store whose manifest.json version differs from currentStoreVersion, removePreviousStore(at:) is invoked to wipe the previous data. It removes storeURL entirely and recreates it as an empty directory — but does not recreate the blobs/ subdirectory. The blobs/ directory was created earlier in init (via Files.createDirectoryIfNeeded(at: blobsURL)), and after removePreviousStore it no longer exists for the rest of the session.

For every blob captured during this session whose compressed size exceeds Configuration.inlineLimit (16 KB by default — i.e. almost any real-world JSON response body), storeBlob falls through to:

try? processedData.write(to: makeBlobURL(for: key.hexString))

The write silently fails because the parent directory is missing. The LoggerBlobHandleEntity row is still saved (with inlineData == nil, no file on disk), and responseBodySize is populated on NetworkTaskEntity. The console then shows:

Unavailable
The response body was deleted from the store to reduce its size. Increase responseBodySizeLimit of the store.

The error message is misleading — responseBodySizeLimit had nothing to do with it; bumping it doesn't help.

Reproduction

Empirical reproduction on iOS 26 simulator with a host app integrating Pulse:

1. Use Pulse 5.1.4 (currentStoreVersion = 3.6.0). Capture a network response > 16 KB compressed. Confirm a file exists at current.pulse/blobs/<sha1> matching the ZLOGGERBLOBHANDLEENTITY.ZKEY row.
2. Upgrade to Pulse 5.2.1 (currentStoreVersion = 3.7.0). Relaunch the app without uninstalling (so the old manifest.json is present). Make another large request. Observe:
   • manifest.json is now version: \"3.7.0\" (migration ran)
   • blobs/ directory exists but is empty
   • ZNETWORKTASKENTITY has the task with non-zero ZRESPONSEBODYSIZE and a non-null ZRESPONSEBODY FK
   • The referenced ZLOGGERBLOBHANDLEENTITY row exists, but ZINLINEDATA is NULL and no file exists at blobs/<hex(ZKEY)>
3. PulseUI shows "Unavailable" for the response body.

Confirmed end-to-end on a real app upgrade path — this is what users will hit when bumping their dependency from 5.1.x to 5.2.x.

Fix

Mirror the cleanup pattern already used by _removeAll() (LoggerStore.swift:862-863) and recreate the blobs/ subdirectory at the end of removePreviousStore:

private static func removePreviousStore(at storeURL: URL) throws {
    try Files.removeItem(at: storeURL)
    try Files.createDirectory(at: storeURL, withIntermediateDirectories: true)
    Files.createDirectoryIfNeeded(at: storeURL.appending(directory: blobsDirectoryName))
}

After applying the fix, repeating step 2 of the reproduction produces a populated blobs/<sha1> file matching the entity row, and the response body renders correctly in PulseUI.

Scope

One-line change. Affects only the migration path — fresh installs already work because the initial Files.createDirectoryIfNeeded(at: blobsURL) in init is never invalidated.