Performance, stability, and terminal compatibility improvements

- Replace VLA stack arrays with heap allocation (hmap, gnx, gny, gnz, saved_data) to avoid stack overflow on large terminal sizes
- Refactor render buffers from per-cell string arrays to flat signed char/int arrays with memset for faster clears
- Batch entire animation frame into a single write() call instead of per-cell printf, reducing syscall overhead
- Pre-compute Blinn-Phong half-vector per-frame for lighting calculations
- Pre-compute ANSI color escape sequences into a lookup table
- Fix volatile qualifier: change term_resized to sig_atomic_t for signal safety
- Process tree walk: walk up multiple levels (not just one) to reliably detect terminal emulator through shell wrappers
- Add detection for alacritty, wezterm, ghostty via environment variables
- Fix SGR sequence parsing in colored logos: properly handle bold (1) and reset (0/22) codes, and truncate at last complete SGR before cursor movement
- Change memory/swap counters from long to long long for systems with >2TB RAM
- Fix floating point casts in disk stats for better precision
- Remove unused variable in gather_shell