Add new model architectures and enhance UI with documentation updates

[luisrguerra]

[luisrguerra]

New models require a new version of llama.cpp.

[rainlizard]

Looks good, well done! I did a quick review with Codex, if they're false positives just tell me and we can ignore them:

---

Findings:

1. [P1] EasyQuantizationGUI.py:548 reports completion after quantization failure.
   If llama-quantize.exe exits non-zero or Popen fails, the exception is logged, but execution continues to Done, re-enables the UI, and plays the success sound. Users can easily miss that the output failed or is incomplete. Return from the except, or track success and only log/play success on a clean return.

2. [P2] EasyQuantizationGUI.bat:33 uses %ERRORLEVEL% inside parenthesized blocks, so the new error handling is stale-expanded.
   In batch files, %ERRORLEVEL% inside the if not exist (...) block is expanded before python -m venv, call activate, and pip install run. Those failure checks will usually see the pre-block value instead of the command’s real exit code. Use if errorlevel 1 (...) or delayed expansion with !ERRORLEVEL!.

3. [P2] convert.py:228 validates ignored tensor names before ignoring them.
   The PR adds keys_ignore, but the max-name-length check still runs across every key before line 243 skips ignored tensors. Any ignored optimizer/extra-state tensor with a long name can fail conversion even though it would never be written. Filter ignored keys before computing name_lengths.