Fix Peak ISO-TP instability during large UDS transfers (#425)

[frankie-zeng]

Fixes #425

Problem

DoCan / ISO-TP communication with PEAK (PCAN) was unstable during large firmware updates (many UDS 0x36 requests, ~2500 bytes each). Symptoms included:

• TP_TIMEOUT_UPPER_READ ("upper layer read timeout") at random points
• PCANTP_STATUS_QUEUE_TX_FULL from CANTP_Write_2016 causing the bus to be closed
• Incomplete receive-queue draining and inconsistent message lifetime handling in _read()

Root causes found

1. Transient TX queue full treated as fatal — _writeBase and _writeTp called close() on PCANTP_STATUS_QUEUE_TX_FULL, tearing down the whole channel during bursty multi-frame TX.
2. Receive queue not fully drained — _read() returned after the first loopback/RX message via setImmediate, leaving later messages (including completed ISO-TP RX) queued until the next event.
3. Multi-frame RX never delivered — when INDICATION_RX progress reached COMPLETED, the code freed the message but never emitted the read event, causing readTp() to time out.
4. Message lifetime bugs — several early-return paths skipped CANTP_MsgDataFree_2016; the INDICATION_RX completed path could double-free.

Changes (Peak only, TypeScript-only)

• Added bounded retry with short backoff for transient write statuses (QUEUE_TX_FULL, LOCK_TIMEOUT) in _writeBase and _writeTp; no bus teardown on these errors.
• Rewrote _read() to drain the full PCAN queue per callback and always free each read message exactly once (try/finally).
• Resolve ISO-TP TX writes only after multi-frame COMPLETED progress; emit ISO-TP RX reads only after multi-frame COMPLETED.
• Extracted testable helpers in src/main/docan/peak/helpers.ts with unit tests in test/docan/peakHelpers.spec.ts.

Verification

• npm run test -- test/docan/peakHelpers.spec.ts — pass
• npm run typecheck:node — pass
• No native module rebuild required

Requested re-test (@ptsiewie)

Please verify on a PCAN dongle with Windows 11:

1. Run the built-in S19 firmware update script (or equivalent repeated UDS 0x36 traffic, ~2500-byte payloads).
2. Confirm PCANTP_STATUS_QUEUE_TX_FULL no longer closes the bus.
3. Confirm full firmware uploads complete without TP_TIMEOUT_UPPER_READ.
4. Spot-check normal single-frame UDS request/response still works.

Thank you for offering to help validate on real hardware — that confirmation would be very valuable for closing #425.