Fix complex support for BatchedDot in C and JAX backends

[ayulockedin]

Description

This PR adds native support for complex64 and complex128 data types to the BatchedDot Op across both the C and JAX backends, resolving the NotImplementedError raised during graph compilation mentioned by @jessegrabowski in issue #1849

Key technical changes:

• Python Graph Layer: Updated make_node type-checking in pytensor/tensor/blas.py to explicitly allow complex types to be built into the graph.
• C-Backend ABI Compatibility: Bypassed strict C++ complex struct type mismatches by defining alpha and beta as raw float[2] and double[2] arrays. This matches the exact memory layout expected by the underlying Fortran ABI for cgemm_ and zgemm_.
• Safe Pointer Arithmetic: Prevented stride-related memory corruption in the C-loop by casting PyArray_DATA to char* before applying NumPy byte strides. This guarantees safe memory traversal regardless of the underlying primitive byte size.
• JAX Backend: Verified that once the PyTensor Python layer permits complex inputs, execution delegates seamlessly to JAX's native complex matmul without requiring an explicit custom dispatch.
• Testing: Added comprehensive complex execution tests (test_batched_dot_complex) to tests/tensor/test_blas.py. Implemented the modern @ matrix multiplication operator in the tests to ensure zero FutureWarning deprecations are triggered in the CI pipeline.

Related Issue

• Closes Batched dot doesn't support complex inputs #1849
• Related to #

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

[ayulockedin]

@jessegrabowski could u take a look at this PR when u have a moment thx :)