ipw2100: Use GFP_KERNEL instead of GFP_ATOMIC in some memory allocation

The call chain is:
   ipw2100_pci_init_one            (the probe function)
     --> ipw2100_queues_allocate
       --> ipw2100_tx_allocate

No lock is taken in the between.
So it is safe to use GFP_KERNEL in 'ipw2100_tx_allocate()'.

BTW, 'ipw2100_queues_allocate()' also calls 'ipw2100_msg_allocate()' which
already allocates some memory using GFP_KERNEL.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Link: https://lore.kernel.org/r/20200722101701.26126-1-christophe.jaillet@wanadoo.fr