Fixes: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=69350

**`IsADirectoryError`**

Fuzzer provided input data can sometimes produce filenames that look
like directories and raise `IsADirectoryError` exceptions which crash
the fuzzer. This commit catches those cases and returns -1 to instruct
libfuzzer that the inputs are not valuable to add to the corpus.

**`FileExistsError`**

Similar to the above, this is a possible exception case produced by the
fuzzed data and not a bug so its handled the same.

Adds a utility function to limit the maximum file name legnth produced
by the fuzzer to a max size dictated by the host its run on.

The fuzzer was having trouble analyzing `fuzz_submodule.py` when using
the `atheris.instrument_imports()` context manager. Switching to
`atheris.instrument_all()` instead slightly increases the startup time
for the fuzzer, but significantly improves the fuzzing engines ability
to identify new coverage.

The changes here also disable warnings that are logged to `stdout` from
the SUT. These warnings are expected to happen with some inputs and
clutter the fuzzer output logs. They can be optionally re-enabled for
debugging by passing a flag o the Python interpreter command line or
setting the `PYTHONWARNINGS` environment variable.