Follow-ups left open by the Jaguar2 gap-channel review (#501) #502

Two Jaguar2 fast_retune gaps found while reviewing #501. #501 does not touch them. Both are state that only the full channel set writes, so a software fast hop keeps whatever the previous channel left.

• 0x838[7:4] 20 MHz DFS-channel CCA tweak is never rewritten by fast_retune. Reproduced on an 8822BU (0bda:b812) on master and on jaguar2: a defined AGC table, fc and RF 0xBE for every 5 GHz channel #501's head. With a 20 MHz RX sweep over 36,117, ch 117 reads 0x838 = 0x86666351 after a full set and 0x86666361 after a fast hop. Two full runs agree with each other. The quick register check is tests/hop_parity_check.sh at BW=20 INIT=36 TARGET=52 (it injects, so pick in-band channels), or an RX-only DEVOURER_RX_SWEEP vs DEVOURER_RX_SWEEP_FULL=1 comparison with DEVOURER_DUMP_CANARY=1.
• A spur_calibration notch left by a full set is never cleared by fast_retune. Found by code reading, not measured yet. A full set onto a spur channel followed by a fast hop off it should leave the notch armed on the new channel.

Done means the fast path writes, or explicitly clears, both, and the fast-vs-full canary comparison shows no differences at 20 MHz across 52..64 / 100..144.