J1939: collected minor findings (PGN normalization, CTS retransmit, claim delay, thread churn)

[dborgards]

Minor findings from section 3 of the repository review, J1939 part. Source: docs/reviews/2026-09-10-repository-review.md (branch claude/repository-review-hpttwc). Line numbers refer to main @ 8ed80f6.

Transport protocol

• A PDU1 PGN from TP.CM is not normalized, so a peer that puts the destination address in the low byte never reaches our TX session.
• A CTS asking for retransmission is answered with an abort instead of being served.
• maxPacketsPerCts == 0 is read as "unlimited"; J1939-21 defines 0 as "hold, send nothing".
• DatagramReceived is raised before the inbox write, on the actor (J1939TpChannel.cs:936-947). ISO-TP already restructured exactly this.

Node

• The 0–153 ms pseudo-random delay before Cannot Claim / re-claim (J1939-81 §4.4.4.3) is missing, so two nodes colliding on an address can livelock in lockstep.
• A second ClaimAddressAsync silently cancels the first.
• Every claim round creates two threads — a full arbitrary-address scan is 240 thread creations.

Docs

• The README maps FR-TP-032 and FR-TP-033 the wrong way round.
• J1939NodeOptions.cs:63 documents a factory overload that does not exist.