[PATCH 0/1] MR5129: ntdll: On x86_64, don't access %gs in signal_start_thread().

[Brendan Shanks (＠bshanks)]

`%gs` is not set until `call_init_thunk()`, this caused a crash on macOS when starting the second thread in a process.

(Paul was curious why this didn't fail on Linux, and found that `%fs` is being inherited from the creating thread.)

Also, adjust TEB accesses in other assembly functions to be consistent with the surrounding code (using registers, not fs/gs).

Fixes a crash on macOS introduced by 7ae488a2bb58501684c6475d4942277b852475fc
("ntdll: Don't hardcode xstate size in syscall frame.")