iOS: classify RR73 sign-off before grid so answered QSOs complete and log

[patrickrb]

Root cause

QsoEngine.classify (ios/FT8AFKit/Sources/FT8Engine/QsoEngine.swift) checked the decoded message's grid field before the RR73/RRR/73 sign-off switch:

if !msg.grid.isEmpty { return .grid(msg.grid) }   // ran first
switch extra { case "RR73": return .rr73; ... }

"RR73" is a valid 4-char Maidenhead-shaped token (R,R fall in the A..R field range; 7,3 are digits), so looksLikeGrid("RR73") is true and the decoder copies it into the grid field (FT8Decoder.buildMessage: if looksLikeGrid(m.extra) { m.grid = m.extra }). A partner's standard closing "RR73" therefore reaches the engine with both extra == "RR73" and grid == "RR73", and classify returned Content.grid("RR73") instead of .rr73.

In processRx, the .rReport stage only completes on .rr73/.rrr/.bye73 content; a .grid hits default: break and returns nil. So the auto-sequencer never completed the QSO, retransmitted our R-report forever, and never logged the contact — for essentially every answered QSO, since RR73 is the most common FT8 sign-off.

Fix

Match the sign-offs before the grid check, so a sign-off is never mistaken for a grid report. This mirrors the two reference ports, which already order it this way:

• desktop desktop/src-tauri/src/qso.rs — sign-off switch first, with a comment describing this exact regression.
• Android GeneralVariables — classifies a token as a grid only when it matches the Maidenhead shape and != "RR73".

iOS was the 1-of-3 divergent port. looksLikeGrid is untouched (the map/distance code relies on its shape test), and behavior is byte-identical for every non-sign-off input (genuine grids, reports, R-reports).

Testing

• Added testRr73WithDecoderGridStillCompletesQso (ported from desktop rr73_with_decoder_grid_still_completes_qso), which feeds the RR73 slot with grid == "RR73" exactly as the decoder emits it live. Fails pre-fix ("RR73 sign-off must complete the QSO"), passes after — verified by reverting only the source reorder. A companion assertion confirms a genuine 4-char grid reply still advances the CQ initiator to send a report.
• The pre-existing testAnswererFullSequenceLogsQso masked the bug by feeding the RR73 slot with grid == "" (not what the decoder emits).
• swift test on the FT8AFKit package: 132/132 green (built on Linux; the Kit is Foundation-only).

Risk

Low. Localized reordering of two lines in one classification function; no DSP, protocol, or interop change. The .bye73 finish path and all other stages are untouched. iOS app-target wiring (LiveEngine) is unchanged; only the Kit engine logic is affected.

🤖 Generated with Claude Code