wifi: mt76: do not hold queue lock during initial rx buffer alloc

In dma init or reset scene, full buffer is needed for all rx rings. Since
this is very time consuming, split the function to perform initial
allocation without holding the spinlock. This avoids causing excessive
scheduler latency.

Signed-off-by: Quan Zhou <quan.zhou@mediatek.com>
Reviewed-by: Shayne Chen <shayne.chen@mediatek.com>
Reviewed-by: Deren Wu <deren.wu@mediatek.com>
Link: https://patch.msgid.link/57c68a7ce1dd9022fa5e06af2c53d6313f30ec83.1731069062.git.quan.zhou@mediatek.com


Signed-off-by: Felix Fietkau <nbd@nbd.name>