Optimize riscv's cpu_fetch_syscall_args()

[trasz]

Summary

Optimize riscv's cpu_fetch_syscall_args(), making it possible
for the compiler to inline the memcpy.

[trasz]

Fix tinderbox.

[trasz]

Better version.

[arichardson]

If anyone is interested, here is some compiler output to show that the latest version avoids a branch compared to the first one: https://cheri-compiler-explorer.cl.cam.ac.uk/z/T4fdEa

[mhorne]

sys/riscv/riscv/trap.c

101

No longer needed.

124

Should be dst_ap ?

[trasz]

sys/riscv/riscv/trap.c

124

Sigh, you're right. It's been manually cherry-picked from another tree.

[trasz]

Make it actually build.

[mhorne]

Looks good. Presumably this passed basic testing in CheriBSD?

[jrtc27]

Yes, it passed our basic CI testing (boot to multiuser, SSH in, run our own cheri(abi)test unit test programs and run the cat kyua tests).

NB: arm and arm64 are the same as riscv currently is, so if we're changing riscv we should really make the same change to those two ports too.