Parallelize SHA-512 userId recovery (fixes #4 timeout)

[official-Cheongae]

Problem

DeviceInfo.recoverUserIdFromSHA512 scanned 0..1_000_000_000 single-threaded with a 10-second timeout. Modern KakaoTalk account numbers are 8–10 digit integers (hundreds of millions), so on a real machine the scan times out before reaching the pre-image and userId() throws — auth then falls back to "auto-detection failed" (issue #4).

Observed on macOS 26.5 / KakaoTalk 26.7.0: single-thread scan never reached the account id within 10s.

Fix

• Shard the search across all cores with DispatchQueue.concurrentPerform.
• Raise the ceiling to 5e9 and extend the deadline.
• A small @unchecked Sendable result holder (guarded by NSLock) carries the first hit and a shared cancellation flag, so every worker stops the instant one finds the pre-image or the deadline passes.

No behavior change to the derivation itself — same CC_SHA512(String(i)) pre-image, just found in parallel.

Verification

• Release build clean on macOS 26 / Swift 6.3.3.
• Confirmed all cores engage during recovery (≈12× CPU on a 15-core M-series).

Note: on KakaoTalk 26.7.0 I still could not decrypt with the plist candidate ids — that looks like a separate key-derivation compatibility issue (cf. #14), out of scope for this PR. This PR only removes the recovery timeout so a valid id in range is actually found.