association: give up joining after repeated max-backoff failures

[geonnave]

Nodes that failed to join (shared-uplink collisions, or a link/fade problem)
kept retrying the same gateway until the 5 s wall-clock timeout, which left a
long formation tail: a few stragglers taking 5-6 s while the bulk joined under 3 s.

This adds an attempt-based give-up. The node counts consecutive join failures
while already pinned at the max backoff window, and after MARI_BACKOFF_MAX_STREAK
(3) of them it gives up and rescans instead of waiting out the timeout. The 5 s
timeout stays as a wall-clock backstop. Because it is counted in attempts it
adapts to slotframe size automatically (unlike the fixed 5 s); worst case is
about 2 s on the huge schedule. The earlier rescan lets a straggler re-sync on a
fresh gateway/channel rather than hammering a possibly-faded link, which pairs
with the beacon/scan channel hopping already on develop.

Validated on a 100-node join storm: the tail tightens noticeably - p100 drops to
~3.8 s (max 4.8 s) from ~5-6 s, with p95 ~3.0 s and a much smaller spread. A
small p95 cost buys a far more predictable worst case.

MARI_BACKOFF_MAX_STREAK is the tuning knob: set too low it risks premature
rescans under normal join-storm congestion; 3 held up well in testing.

Also included: a testbed LED-color mapping so the current nRF5340-DK gateway
lights the node green (same known-gateway convention already in board.c).

[geonnave]

Tested with 100 nodes and results are amazing! Mean join time for 100 nodes, 10 meters apart from gateway, is 3.8 seconds 😎