fix(firmware): track the quiet floor continuously instead of latching at boot

[clonea1]

calibration_update() averages the first EDGE_CALIB_FRAMES frames and latches
the result for the lifetime of the boot. That cannot be made to work: the node
is calibrating in the seconds right after a human walked over and plugged it
in, so the window is contaminated by construction, and nobody can leave a room
fast enough to fix it. Whatever the room looked like during that one minute
becomes "normal" permanently, and a node commissioned during a busy afternoon
stays desensitised for weeks.

A leaky minimum needs no clean room and no cooperation. It drops instantly to
any new low, so it finds the quiet by itself whenever quiet next happens --
minutes after a bad start rather than never. It climbs only slowly, so a
genuine change in the room is eventually adopted while a person standing still
for a while is not mistaken for the new floor.

if (motion < s_floor) s_floor = motion;        /* descend at once */
else                  s_floor *= EDGE_FLOOR_LEAK;

The warm-up is kept, but only to seed a plausible starting value rather than to
latch one.

Measured support for the shape, on a nine-node ESP32-C6 fleet: the room's true
floor sits about 14x below where household animals register, and the settling
period after human activity runs a couple of hours. A tracker that descends to
real quiet and climbs at well under a percent a minute resolves that; an
average over one arbitrary minute cannot.

Separated by hand from a larger local change rather than cherry-picked -- the
same file also carries 256-bin HE20 subcarrier support and a vitals-slots
packet, which are unrelated and are not included here.

---

Rebased onto current main before opening: staged before today's seven merges, so it needed replaying to avoid reading as a revert of them. Clean rebase, no files deleted.