[bsatrom]

Summary

CommandTask processed only a single command per poll: it acquired the I2C
mutex, called notecardGetCommand() once, then went back to sleep. Because
notecardGetCommand() returns exactly one note per call
(SongbirdNotecard.cpp:585-682), a backlog of queued commands drained at one
command per poll interval — up to 60s apart in storage mode. A user who queued
several commands could wait minutes for the last one to execute.

Fix

Wrap the get-execute step in a bounded drain loop: keep calling
notecardGetCommand() until command.qi is empty or
COMMAND_DRAIN_MAX_PER_POLL (8) commands have been processed this poll. The
I2C mutex is re-acquired and released per command so the bus is not held for
the whole drain, and the cap prevents a flood of commands from monopolising the
bus or starving other CommandTask work.

• src/core/SongbirdConfig.h — new COMMAND_DRAIN_MAX_PER_POLL constant with rationale.
• src/rtos/SongbirdTasks.cpp — drain loop in CommandTask, replacing the single-shot get.

Tests

Adds test/test_command_drain/ native unit tests exercising the drain-loop
control flow against a mock command source (get-until-empty, budget cap,
empty-queue probe, exact-budget, positive-constant). Avoids the FreeRTOS/I2C
hardware dependencies while covering the same branch logic.

Verification

• pio run — both envs pass:
  • blues_cygnet SUCCESS
  • cygnet_debug SUCCESS
• pio test -e native — 96 test cases succeeded (incl. new test_command_drain).

Closes review finding M7