fix(ra): let a failed disc hash explain itself in the log

[androosio]

Description

A user reported that no PlayStation 2 disc in their library matched RetroAchievements and sent an app.log. The only trace of the problem in it was the pass's own summary:

RA re-match (hashMissing): 50 candidate ROMs
RA re-match finished: 50 processed, 38 hashed, 28 matched, 12 skipped

Everything explaining those 12 was lost. Hashing runs under compute(), and a spawned isolate gets its own copy of the LoggerService singleton which never runs init() — so its output is ConsoleOutput alone. CHD: could not open, RA disc: no usable track, RA disc: could not locate primary executable all went to logcat, which a user does not have. The isolate cannot simply open the log file too: two isolates holding a FileOutput on one path interleave writes and race the rotation rename, which is exactly why the secondary display's engine deliberately skips init().

So the isolate hands its lines back instead. LoggerService gains a per-isolate capture (startCapture / takeCapture / replayCaptured), _generateHashForSystemIsolate returns what it collected alongside the hash, and the main isolate replays it into the file-backed logger. Replaying goes through log() like any other line, so RedactingPrinter still strips credentials before anything is written. A capture is capped at 64 lines — it exists to explain a single failure, and an unbounded list in a background isolate is a leak waiting for a pathological image.

The skip count now says why, per run and across the library:

RA re-match finished: 50 processed, 38 hashed, 28 matched, 12 skipped (error 12)
RA re-match: 12 ROM(s) parked (error 12)

The second line is what finally gives getRaHashSkipCounts() a caller, and it earns its place: a pass over a settled library walks none of the parked ROMs, so on the run where a user notices the gap the per-run line reads as a clean sweep.

No behaviour changes — this only makes an existing failure legible. The underlying PS2 report is still open and gets its own PR once a log names the cause.

Steps to Verify

Preconditions: any platform, a library with at least one disc-based system, RetroAchievements signed in.

1. Put a file the disc reader cannot open where a PS2/PSX ROM would go — a .chd of zeroes is enough (head -c 8192 /dev/zero > "Broken (USA).chd").
2. Scan so it is picked up, then run the RetroAchievements match pass (Tools → re-match, or launch with match-on-startup enabled).
3. Open app.log in the user-data folder.
4. Before this change the file holds only the ... N skipped summary. After it, the two warnings that used to reach logcat only are in the file:

   [W]  CHD: ChdException(open): .../Broken (USA).chd
   [W]  RA disc: could not open .../Broken (USA).chd
   

5. The summary line now carries a reason breakdown, followed by the library-wide parked count.

Tested on

• Android — device:
• Linux — device: CachyOS desktop (flutter test, plus a real failing container through RaDiscHash.compute)
• Windows
• macOS
• Not runtime-tested — why: Android not yet exercised. The capture is platform-independent (it sits in LoggerService.log, above every I/O path), and the isolate boundary it crosses is the same on every platform, so the risk of a device-only difference here is low. Happy to smoke it on the Thor before merge if preferred.

Checklist

• dart format . — no diff
• flutter analyze — clean (no errors or warnings)
• flutter test — all passing (1423)
• Tests added or updated — test/logger_isolate_capture_test.dart, 8 cases: per-level tagging, the error object, the 64-line cap, take-clears-capture, replay round-trip, malformed lines, and an unknown tag from a newer build
• No secrets, credentials, or personal data in the diff (including tests and fixtures)
• New assets have compatible licenses — n/a