Update docs: challenges.md (training bias + scene text approach), vam/multimodal/trainers docs sync

Revert camera prompt changes; add NaN guard and CUDA sync; add engineering challenges doc

refactor: rename VLM configs, add VLM SFT configs, and improve model loading

refactor: reorganize dataset into lm/vlm/vam subdirectories

docs: update README and docs for model_type rename, RoPE guard fix, config paths

reorganize configs into subdirs lm/vlm/vam, remove runs/, add interview docs for pretrain & full_sft

cleanup: remove deprecated doc files, update src training utils

docs: move tokenizer.md into interview/

docs: add comprehensive tokenizer deep dive doc (562 lines)

refactor(runs): rename lm/vlm/vam.sh to train_*

refactor(configs): flatten configs/model/ into configs/

refactor(docs): restructure docs with modular organization

Replace generic README.md doc names with descriptive index/overview names

Restructure docs to mirror src/ module layout

Restructure docs index into categorized sections, move architecture overview into architecture/

Update docs for tokenizer training script and checkpoint save path

Add modular development docs under docs/ for learning and interviews