bcm2835_sdhci: improve busdma usage

[kevans]

Summary

Currently, the busdma(9) tag is created with maxsize=512, nsegs=512, magsegsize=512, meaning (if I interpret this correctly) we will only ever load exactly one segment despite clearly expecting at most two. As such, we end up remapping like it's going out of style every 512 bytes.

I suspect this will have a decent impact on mmc performance.

Test Plan

Ask other people to test, because I'm still working out how to cope with addressing restrictions on the rpi4 and don't have functional sdhci+dma at the moment.