apps-sandbox/dotbot: drive the robot through drv/dotbot_control

[geonnave]

The sandbox app's control glue now lives in DotBot-libs' hardware-free control
core (drv/dotbot_control) instead of being reimplemented inline in
apps-sandbox/dotbot/main.c. main.c drops from 1083 to 554 lines. The app
keeps its own mailbox, qdec, and localization polling via
db_control_fix_due, writes motors through out.write, sends advertisements
through out.advertise, and keeps the RGB LED command and bench hooks.

Behaviour differences from the previous inline implementation:

• Deadman now counts scheduler ticks starting from when the mailbox applies a
  command, rather than from when the command was received.
• Motors are written once at the end of each tick rather than incrementally
  during it.
• The minimum tx interval is re-read per advertisement, same as before.

Submodule

The dotbot-libs submodule points at the merge commit of
DotBots/DotBot-libs#38 on main.

Bench status

Validated on 3 real robots running this build against a 1.24.0 control
baseline running on a 4th. Matching 1.24.0 behaviour: single- and
multi-point waypoints with final heading, max speed, stop/abort, deadman,
advertisements, and LED.

Not yet passing on the previous build: the LOST re-anchor step (lifting a
robot mid-batch) did not recover. The robot reseeded while still in the hand
(its braked wheels passed the "wheels still" test during a slow carry), spun
in the air, and failed with NO_HEADING. The dotbot-libs pointer now carries
three fixes from DotBot-libs #38: a LOST robot reseeds only once its fixes
hold still for 0.5 s; wheels that turn while the fixes stay put drop the pose
back to LOST, so steering brakes instead of spinning in the air; and a
NO_HEADING spin that ends without a heading rests in HOLD and retries (up to
3 times) before failing.

A second lift test on that build showed the retries spinning the wheels 3-5
times in the hand. The pointer now also carries the follow-up: a pose lost on
a free spin (wheels turning while the fix stays put) fails the batch at once,
braked, with no retry, so a held robot gives one short spin burst and stops.
The retries remain for spins that moved the fix. Before the change, the same
robot on the previous build ran 180 s of untouched laps at 150 mm/s with no
loss of tracking, which rules out a false free-spin trigger on the floor.
This build, flashed on that robot, ran 180 s of untouched laps just as
cleanly: 4 batches ARRIVED, no loss of tracking after the initial heading
spin. The lift mid-batch step itself still needs re-running at the bench.

Build: SES 7.22, sandbox-dotbot-v3 Release, 0 warnings.

File	+	-
apps-sandbox/dotbot/main.c	+150	-679
2 small files: apps-sandbox/applications.emProject, dotbot-libs	+2	-2
Total, 3 files	+152	-681