fix(interp): don't fseek(NULL) on M99 in MDI

[grandixximo]

Problem

Entering M99 as an MDI command crashes milltask with SIGSEGV. Gmoccapy and QTdragon die; AXIS happens to survive. Reported in #4094.

Backtrace:

#5  fseek () from libc.so.6
#6  Interp::loop_to_beginning (...) at emc/rs274ngc/interp_o_word.cc:526
#7  Interp::convert_stop (...) at emc/rs274ngc/interp_convert.cc
#8  Interp::execute_block (...)
...
#12 emcTaskPlanExecute (command="m99", ...)

Cause

Task always enables loop_on_main_m99 (emctask.cc), so M99 in the main program takes the Fanuc endless-loop branch in convert_stop(), which calls loop_to_beginning() -> fseek(settings->file_pointer, 0, SEEK_SET). In MDI there is no open g-code file, so file_pointer is NULL and fseek(NULL) segfaults.

Fix

• Require file_pointer != NULL to take the M99 loop branch. In MDI, M99 now falls through to the existing M2/M30 program-end path, which already null-guards file_pointer. This matches AXIS-style "end" behavior.
• Add a defensive NULL check in loop_to_beginning().

Testing

Built against current master. Verified on Gmoccapy: M99 in MDI no longer crashes.

Fixes #4094