so3: terminate the faulting user process on an EL0 abort instead of panicking

[daniel-rossier]

Problem

A synchronous abort taken from EL0 — a fault in the running user process, e.g. a NULL-pointer dereference — used to fall through trap_handle()'s default path to trap_handle_error() + kernel_panic(), bringing down the whole kernel (infinite loop, IRQs off). A single buggy user program hangs the system.

Fix

Handle EL0 aborts (ESR_ELx_EC_DABT_LOW / ESR_ELx_EC_IABT_LOW) in a new user_fault() (arch/arm64/traps.c):

• report which process faulted and where (process "<name>" (pid N) faulted at address 0x… (pc=0x…, ESR=0x…));
• sys_do_exit_group() to terminate only that process; the parent's wait4() reaps it and the shell reprints its prompt.

Kernel-mode (EL1) aborts use a different exception class (DABT/IABT "current EL") and still reach the fatal path, so real kernel bugs still panic. Scoped to !CONFIG_AVZ (AVZ keeps its mmio-decode / s3c_crash path).

Verification

Built + run in QEMU (virt64, standalone). A user process that dereferences a NULL pointer now prints e.g.:

[SO3] process "/crashme.elf" (pid 3) faulted at address 0x8c (pc=0x400238, ESR=0x92000046) - terminating it.
/ %

and the shell stays alive (ls afterwards works), instead of the previous ### ERROR CAUGHT … kernel_panic() hang.