Use local RNG in MultiTensor tests

[fzlzjerry]

Summary

• give every randomized MultiTensor test a fresh, deterministically seeded numpy.random.Generator
• pass that generator into the tests explicitly and replace all 29 legacy global RNG calls in the module
• keep the suite-wide seed in place for modules that have not yet been migrated

Rationale

This is an incremental part of #1060. It makes the ten randomized tests in _multitensor_test.py isolated and reproducible without changing random behavior elsewhere in the suite.

Validation

• check/all --only-changed-files origin/main (2287 passed; formatting, Pylint, mypy, incremental coverage, ShellCheck, and notebook formatting all passed)
• representability suite (44 passed)
• MultiTensor tests with pytest-randomly seeds 1, 42, and 4294967295 (10 passed for each seed)
• MultiTensor tests with legacy np.random.random patched to raise on use (10 passed)

[gemini-code-assist]

Code Review

This pull request refactors the test suite in _multitensor_test.py to use a deterministic random number generator. It introduces a pytest fixture rng utilizing np.random.default_rng(0) and updates all test functions to use this fixture instead of calling np.random.random directly. There are no review comments to process, and I have no additional feedback to provide.