Fix memory corruption in headless (nographics) build

[laurensWoon]

Summary

Any procedure containing 2+ turtle-movement calls (fd/back) crashes a --disable-wx (nographics) build with EXC_BAD_ACCESS.

Root-caused with lldb: nographics.h defines GR_SIZE as 1 ("a dummy graphics header file for computers without graphics"), so graphics.c's char record_buffer[GR_SIZE]; is a single byte -- but safe_to_save() (called from save_line()/save_move(), not gated by any nographics-specific stub) writes a full 8-byte pointer into it on the very first call, silently corrupting whatever global follows record_buffer in memory (observed: AllowGetSet). The corruption doesn't crash immediately -- it surfaces later, on some unrelated evaluator lookup that happens to read the now-garbage pointer, which made this confusing to trace back to its actual cause (looks like the second movement command crashes, not the first).

fd/back are the only movement primitives that go through forward_helper() (and thus save_line()/save_move()); rt/lt don't call it at all, which is why a procedure with only turns never triggered this.

Fix

Make safe_to_save() an unconditional no-op for headless builds, mirroring the existing HAVE_WX-only early-return already in draw_turtle() a few lines above it. There's no window to ever redraw in a headless build, so recording draw operations for later replay serves no purpose there -- bumping GR_SIZE instead would only move the overflow further out, not fix the underlying mismatch.

Also added the equivalent early-return to draw_turtle() itself for headless builds (previously only guarded for HAVE_WX) -- turned out not to be this bug's cause once traced with lldb, but it's the same class of issue (recursing through forward() with nothing to actually draw) and worth closing off regardless.

Test plan

Verified with koch 150 4 (256 nested fd/lt/rt calls from a self-recursive procedure) -- previously crashed immediately, now completes cleanly with a correct final turtle position.

Related to (does not fully explain) #245, which covers a distinct, not-yet-root-caused issue in the wx build.

[benjaminpjones]

instead of checking a list of platforms, would it make sense to gate on the memory constraint we care about? In other words

#if GR_SIZE <= GR_SLACK

where GR_SLACK is a sufficient buffer for safe_to_save (in the ballpark of 300)

[benjaminpjones]

cs (clearscreen) also writes to record_buffer. I believe it would require a similar fix.

ucblogo-code/graphics.c

Line 775 in ebb9c42

record[record_index] = FINISHED;