Fix: reuse GDB packet buffer for m/M-packets instead of VLA

[ALTracer]

Detailed description

• This is not a new feature, but still a potentially dangerous change. A sort of continuation of PR2014.
• The existing problem is gdb_main FSM performing VLA for m/M-packets.
• This PR solves it by reusing half of gdb packet buffer for (un)hexify operations.

Tested on BMDA (env CC=clang, -Db_sanitize=address) and on blackpill-f411ce against GD32VF103. Dump and compare-sections seem to match, GDB receives 512 byte (1024 hex-encoded) packets.
g/G-packets still depend on VLA because BMP supports multiple architectures with different amounts of general-purpose registers. {armv7-m 20*4, armv8-m.main 24*4} + fpv4-sp-d16 33*4, armv7-a/r 17*4 + VFPv3 33*4, armv8-a 31*8, rv32i 33*4, rv32e 17*4, rv64i 33*8. IMHO a static regcache of 512 bytes could be enough (with build-time static assert that no arch needs more), but it's not visibly broken now, so not touching.

Your checklist for this pull request

• I've read the Code of Conduct
• I've read the guidelines for contributing to this repository
• It builds for hardware native (see Building the firmware)
• It builds as BMDA (see Building the BMDA)
• I've tested it to the best of my ability
• My commit messages provide a useful short description of what the commits do

Closing issues

Closes #2042. @perigoso

[dragonmux]

LGTM, please rebase on main and we'll get this merged. Thank you for the contribtion! We quite like this way of working round the VLA problem - it's really a nice solution.