deps: cap transformers<5.13 — 5.13.0 breaks mlx-lm import, killing every fresh install at model load (#136, #135)

transformers 5.13.0 changed AutoTokenizer.register to require a config
class as the key; mlx-lm 0.31.x registers by name ("NewlineTokenizer"),
so `import mlx_lm` raises AttributeError('str' has no '__module__')
before any model can load. Every venv resolved after 5.13.0 shipped is
dead on arrival — three independent reporters in 24h (#136 + 2 dupes,
#135 is timing-consistent), and we hit the identical crash ourselves on
2026-07-03 and pinned the local venv manually (this makes the fix
survive the app's fingerprint force-reinstall, which the manual pin did
not). Marker-gated like the sibling mlx pins.

Same content as PR #137 by @davidtai (verified: bisect 5.12.1 good /
5.13.0 bad matches our own finding). Wheel metadata verified:
Requires-Dist: transformers<5.13 present in the built wheel.

Drop the cap once mlx-lm registers the 5.13-compatible way
(ml-explore/mlx-lm#1465).

Author: youssofal

pyproject.toml

@@ -16,6 +16,12 @@ dependencies = [
   "huggingface-hub>=0.36",
   "mlx>=0.31,<0.32; sys_platform == 'darwin' and platform_machine == 'arm64'",
   "mlx-lm>=0.31,<0.32; sys_platform == 'darwin' and platform_machine == 'arm64'",
+  # transformers 5.13.0 changed AutoTokenizer.register to require a config
+  # class as the key; mlx-lm 0.31.x still registers by name
+  # ("NewlineTokenizer") and crashes at import, killing every fresh install
+  # at model load (#136, #135; PR #137 by @davidtai). Drop the cap once
+  # mlx-lm registers the 5.13-compatible way (mlx-lm PR #1465).
+  "transformers<5.13; sys_platform == 'darwin' and platform_machine == 'arm64'",
   "nanobind>=2; sys_platform == 'darwin' and platform_machine == 'arm64'",
   "numpy>=2",
   "pydantic>=2",