Add stream transpose_3d implementations

[hadronic-arvind]

Summary

Fix transpose_3d for stream I/O in the Quartus and Catapult backends.
Add stream-mode transpose helpers that unpack the input stream, apply the 3D permutation mapping, and repack the output stream.
Add a regression test for a Keras Permute((3, 1, 2)) model using io_stream.
Make the pyproject-fmt pre-commit hook explicitly install tomli so the Python 3.8 format job does not fail on the hook environment.

Fixes #712.

Tests

Reproduced the Quartus failure before the fix in a clean local clone:

error: no matching function for call to 'transpose_3d'
candidate function template not viable: no known conversion from 'stream<input_t>' to 'input_t *'

After the fix:

PATH="/Users/arvindtawker/Documents/Codex/hls4ml/.agents/bin:$PATH" /Users/arvindtawker/miniconda3/envs/hls4ml-py310/bin/python -m pytest -q test/pytest/test_transpose_concat.py::test_stream_3d_permute_pack_mismatch --override-ini addopts=
4 passed

Additional verification:

PATH="/Users/arvindtawker/Documents/Codex/hls4ml/.agents/bin:$PATH" /Users/arvindtawker/miniconda3/envs/hls4ml-py310/bin/python -m pytest -q test/pytest/test_transpose_concat.py -k 'not oneAPI and not oneapi' --override-ini addopts=
14 passed, 4 deselected

PATH="/Users/arvindtawker/Documents/Codex/hls4ml/.agents/bin:$PATH" /Users/arvindtawker/miniconda3/envs/hls4ml-py310/bin/pre-commit run --all-files
Passed

[vloncar]

Has this been validated on hardware?