fix(ti99): correct TI-99/4A config (speechsyn topology, Game Focus, on-board controls)

[agiacalone]

Summary

Corrects the TI-99/4A (ti99) configuration: the right MAME command, keyboard input that actually reaches the core, and full on-board (handheld) controls. Validated on a TrimUI Brick Hammer (Allwinner A133P, aarch64), scarab 20260511, from a clean deploy of this branch.

Scope (honest): these are configuration-layer fixes — necessary and correct, but not sufficient alone. On scarab 20260511 the stock mame_libretro core itself SIGSEGVs running ti99_4a even with the corrected command — tracked separately in #85 (and same family as #81). Playable TI-99 also needs a working MAME core; validated here against MAME 0.287.

Changes

1. libretroMAMEConfig.py + mameGenerator.py — correct TI-99 slot topology. -ioport:peb:slot3 speech is invalid (speech is not a PEB-slot card; the synth is speechsyn on the console ioport, PEB chains through its extport). lr-mame failed to load the machine on the bad option. Fixed so speech + 32K RAM coexist.
2. libretroConfig.py — input_auto_game_focus = '2' (general fix). Without Game Focus, RetroArch never forwards keyboard input to the core, so no MAME computer system can receive keys in-game. Detect mode enables it only for keyboard cores (other systems keep hotkeys). Benefits every MAME computer system (apple2, bbc, coco, atom, electron, fm7, … and ti99); cherry-pickable on its own. Likely also helps Conflicting/buggy hotkey behavior on MAME 2003-plus RetroArch core (and possibly other MAME cores) #82.
3. board/fsoverlay/usr/share/evmapy/ti99.keys — default on-board control map. The libretro MAME core routes the pad only to MAME joystick inputs, never the keyboard matrix; evmapy (no-grab) emits real KEY_* via uinput → reaches the core once Game Focus is on. d-pad = joystick, A/B = fire, L1/R1/L2/R2/X = 1–5, Select = 0, hotkey+Start = exit.

Testing (TrimUI Brick Hammer, A133P)

Deployed this branch to real install paths, restored the stock core, stripped all runtime overrides, launched a-maze-ing:

• Stock scarab core → SIGSEGV on the correct -ioport speechsyn command (cmdfile + fresh .pyc confirmed) → core defect (TI-99/4A: mame_libretro core crashes (SIGSEGV) at launch on scarab 20260511 (ARM/A133P) — core defect, not config #85), not config.
• With MAME 0.287 + this config: boots, speech runs, digits work from the default evmapy map, and the d-pad moves in the maze via the joypad path with no per-machine cfg — config side is complete.
• All four ti99 speech/32K command combinations validated via -listslots.

Related

• TI-99/4A: mame_libretro core crashes (SIGSEGV) at launch on scarab 20260511 (ARM/A133P) — core defect, not config #85 — the TI-99 core crash this PR's config does not fix (the remaining blocker).
• MAME 2010 (MAME 0.139) RetroArch Core crashes at launch #81 — MAME 2010 core crash; same core-regression family.
• Conflicting/buggy hotkey behavior on MAME 2003-plus RetroArch core (and possibly other MAME cores) #82 — MAME hotkey/input; change 2 (Game Focus) is in that area.