Print planned blocks from the grbl thread, not the simulator thread

[DavidLDawes]

Problem

The simulator occasionally segfaults, most often just after startup. Every crash dump I captured (7 from real runs, 5 of them from a plain master build) has the same stack on the simulator thread:

#0 plan_get_recent_block ()
#1 printBlock ()
#2 sim_loop ()
#3 main ()

printBlock() reads the planner through plan_get_recent_block() (block_buffer.head->prev), but it's called from grbl_per_byte() on the simulator thread. That call is unconditional whenever no client is connected, which includes startup. Meanwhile plan_reset() on the grbl thread malloc()s the block ring and relinks it. The simulator thread can therefore read head before prev is linked, get uninitialized heap as a non-NULL block pointer, and crash dereferencing it.

Fix

The printBlock() call moves into sim_process_realtime(), the grbl.on_execute_realtime hook that already runs on the grbl thread on every main-loop pass. The planner is then only read by the thread that modifies it. The existing rule of not printing blocks to a live console in socket mode is unchanged.

Evidence

400 launches of grblHAL_sim -p … -n -t 0 -b <file>, counting segfaults in the first 250 ms:

Build	Segfaults
master 91eda77	2 / 400
master + this commit	0 / 400
A build with the same printBlock() code plus #25	11 / 400 (race timing varies between builds)
That build + this commit	0 / 400

• Block output is unchanged: socket mode with -b file logs one line per move with the same step counts; stdin mode prints blocks to stdout and exits cleanly on ^F.
• Downstream: in a project whose CI drives the simulator over -p on every push, this crash was the cause of intermittent CI failures, and it hasn't recurred with this change.

Notes

• Independent of Make the driver's atomic flag helpers actually atomic #25 (the atomic flag helpers): this commit applies directly on master and touches different code.
• A separate issue in core, not addressed here: AddressSanitizer also flags settings_write_build_info() in core's settings.c. It copies sizeof(stored_line_t) bytes from BUILD_INFO, a string literal that can be a single byte, an over-read on first boot.
• Built and tested with gcc on Linux only.

🤖 Generated with Claude Code