bump(main/python-numpy): 2.1.1 and update revdeps

Update to numpy 2, and update reverse dependencies:

- matplotlib from 3.9.1 to 3.9.2
- python-contourpy from 1.2.1 to 1.3.0
- python-numpy from 1.26.5 to 2.1.1
- python-onnxruntime from 1.19.0 to 1.19.2
- python-scipy from 1.14.0 to 1.14.1
- python-torch from 2.1.2 to 2.4.1
- python-torchaudio from 2.0.1 to 2.4.1
- python-torchvision from 0.15.1 to `0.19.1
- Revbump: python-pyarrow, inkscape and opencv

Some of these packages does not currently build before this PR. Getting these updated and buildable is a nice step to prepare for [python 3.12](https://github.com/termux/termux-packages/pull/18078).

Fixes #21059, #21110, #21194 and #21284.

Co-authored-by: Chongyun Lee <45286352+licy183@users.noreply.github.com>

x11-packages/python-torch/torch-csrc-autograd-python_function.cpp.patch

@@ -0,0 +1,17 @@
+Fix 32-bit builds, which otherwise fail with:
+
+	python_function.cpp:798:15: error: format specifies type 'long' but the argument has type 'int' [-Werror,-Wformat]
+  	796 |               "save_for_backward can only save variables, but argument %ld is of "
+
+diff -u -r ../cache/tmp-checkout/torch/csrc/autograd/python_function.cpp ./torch/csrc/autograd/python_function.cpp
+--- ../cache/tmp-checkout/torch/csrc/autograd/python_function.cpp	2024-10-03 18:00:30.204986580 +0000
++++ ./torch/csrc/autograd/python_function.cpp	2024-10-04 15:06:58.912498321 +0000
+@@ -795,7 +795,7 @@
+           throw torch::TypeError(
+               "save_for_backward can only save variables, but argument %ld is of "
+               "type %s",
+-              i,
++              (long) i,
+               Py_TYPE(obj)->tp_name);
+         }
+       }