Add public GC_start_mark_threads() to allow parallel marker in fork child

Add public GC_start_mark_threads() to allow parallel marker in fork child

* include/gc.h (GC_start_mark_threads): New API function prototype.
* misc.c (GC_start_mark_threads): New API function definition (empty)
if THREADS but not PARALLEL_MARK or not CAN_HANDLE_FORK.
* pthread_support.c (available_markers_m1): New static variable or
macro (redirecting to GC_markers_m1) depending on CAN_HANDLE_FORK
(only if PARALLEL_MARK).
* win32_threads.c (available_markers_m1): Likewise.
* pthread_support.c (start_mark_threads): Redirect to
GC_start_mark_threads (and decorate with GC_API/GC_CALL) if
CAN_HANDLE_FORK (and PARALLEL_MARK); skip start if parallel markers
are disabled or already started if CAN_HANDLE_FORK; iterate up to
available_markers_m1 (instead of GC_markers_m1); always set
GC_markers_m1 value.
* win32_threads.c (start_mark_threads): Likewise.
* pthread_support.c (GC_thr_init): Set (and test) available_markers_m1
value instead of GC_markers_m1/GC_parallel (only if PARALLEL_MARK).
* win32_threads.c (GC_thr_init): Likewise.
* tests/test.c (NO_TEST_HANDLE_FORK): Do not define if
TEST_HANDLE_FORK.
* tests/test.c (run_one_test): Invoke GC_start_mark_threads (and
additional GC_gcollect) in forked child (only if THREADS but not
NO_TEST_HANDLE_FORK); do not call tiny_reverse_test if not THREADS.
* win32_threads.c (start_mark_threads): Add assertion that the caller
is not holding the allocation lock (to match that in
pthread_support.c) if GC_PTHREADS_PARAMARK.