Bazel builds should be hermetic, but they're not

[mhucka]

Describe the issue

The Bazel configuration files do no install the Python requirements, which means it's up to the user to do pip install requirements.txt and dev-requirements.txt prior to running Bazel commands. If they don't, they may get an unobvious error message such as this:

ERROR: Analysis of target '//tests:simulator_testfixture' failed; build aborted: Problem getting numpy include path.
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'numpy'

We need to make qsim's Bazel files use rules_python and install the Python requirements.

What version of qsim are you using?

0.22.0.dev0

How can the issue be reproduced?

No response

[kris524]

Hello, I had a look at the Bazel setup and wanted to ask about the TensorFlow workspace initialisation
In the WORKSPACE, the TensorFlow targets imports numpy from the system environment. It requires it to be pre-installed. This makes it difficult to make bazel builds truly hermetic. Should we change how we handle the tensorflow targets (tf_workspace*() calls)? Is the TF Quantum compatibility setup still a hard requirement?