xae(4), axidma(4): rewrite DMA operation.

Due to performance constraints on a synthesized CHERI RISC-V core,
remove usage of xdma(4) scatter-gather framework. Instead, provide
a minimalistic interface between two drivers.

This increases performance ~4-5 times.
Tested using scp(1) and nc(1) on Codasip Prime.

Sponsored by: CHERI Research Centre
Differential Revision: https://reviews.freebsd.org/D53932