fix: open job-instance.pickle in binary mode in LSF contrib (#3330)

[1102Aryan]

Description

lsf_runner.py opened job-instance.pickle in text mode ("r"), which fails on
Python 3 because pickle.load() requires a binary stream. This PR:

• luigi/contrib/lsf_runner.py: open the pickle file with "rb" instead of "r"
• luigi/contrib/lsf.py: write the pickle file with "wb" instead of "w" in
  _dump(), and use bytes literals in the __main__-module .replace() call,
  since pickle.dumps() returns bytes on Python 3

Motivation and Context

Fixes #3330. Any LSF task run through lsf_runner.py on Python 3 crashes with
TypeError: a bytes-like object is required, not 'str' (or UnicodeDecodeError,
depending on pickle protocol and platform) before the job's work() method can run.

Have you tested this? If so, how?

I reproduced the reported error on master with a script that pickles a job object
and calls lsf_runner.do_work_on_compute_node() — it fails exactly as described in
the issue. With this fix, the same round-trip succeeds and the job's work() method
runs, including the __module__ == "__main__" dump path. The existing LSF tests
(test/contrib/lsf_test.py) and ruff lint checks also pass.

I have also included a unit test that round trips the pickle through _dump() and lst_runner.do_work_on_compute_node()