memory fault in alien code on non-x86 poisons GC

[Alastair Bridgewater]

Found while reviewing a fix for bug 309068: lisp_memory_fault_error() unconditionally uses fake_foreign_function_call(), exposing the interrupt context to the GC to be scavenged. But if the fault occurs from alien code then the register set isn't our usual partitioned register set and we must NOT use fake_foreign_function_call() (see interrupt_handle_now() for saner handling of this).

Not a recent regression: This has been the case for at least nine years now.

Setting importance to Low because: Nobody has reported this as an issue in most of a decade, and it's for faults in alien code. Debugging alien code isn't a use-case that SBCL has, historically, particularly cared about.