Fix local package detection for packages without top_level.txt

[datatalking]

Adds fallback detection for .dist-info directories that lack top_level.txt (e.g., scipy, seaborn). Parses package name and version directly from the directory name when top_level.txt is missing.

Fixes #374

[StantonMatt]

I checked this locally because #374 has a long trail of reports around packages like scipy/seaborn missing top_level.txt, and the fallback does cover that core shape.

With a synthetic scipy-1.9.3.dist-info directory and no top_level.txt:

origin/master packages []
origin/master import local []
pr-528 packages [{'name': 'scipy', 'version': '1.9.3', 'exports': ['scipy']}]
pr-528 import local [{'name': 'scipy', 'version': '1.9.3', 'exports': ['scipy']}]

So the intended local-detection path works in that direct repro. The branch does need a bit more test work before I would call it ready, though. Full local discovery currently fails:

python -m unittest discover
FAILED (failures=1)

The failing test is tests.test_pipreqs.TestPipreqs.test_init_local_only. In my environment, --use-local now writes:

beautifulsoup4==4.14.3
docopt==0.6.2
ipython==8.18.1
pyflakes==3.4.0

The new beautifulsoup4 entry comes from the existing bs4 import mapping plus the new fallback seeing beautifulsoup4-4.14.3.dist-info without top_level.txt. That may be intended behavior, but the test expectation has not been updated, and there is not yet a committed regression test for the scipy/seaborn-style no-top_level.txt case.

Other checks passed:

python -m py_compile pipreqs/pipreqs.py tests/test_pipreqs.py
python -m flake8 pipreqs/pipreqs.py tests/test_pipreqs.py
git diff --check origin/master...HEAD

GitHub reports no check runs for this PR. I think the next step is to add a small fixture/regression test for a .dist-info directory without top_level.txt, and either update test_init_local_only for the now-detected beautifulsoup4 output or adjust the fallback if that output is not desired.

[hernot]

@datatalking, @StantonMatt: Just some thoughts - I guess you thought your self about and had good reasons not to go this route.
According Python Packaging Guide METADATA is the only mandatory file in *.dist-info diretories. It has to contain package name and version.
So how about the following METADATA centered approach in replacement for simple "fallbacking"?

1. parse the METADATA file in all *.dist-info directories for name and version fields and add both to list of local pacages
2. check if top_level.txt is present in *.dist-info directory and add its content to package list as done now

Edit

I checked pyserial - a package i know is to be imported as serial instead of pyserial -> it has a top_level.txt file in its .dist-info directory. Whereas eg. numpy and scipy, where package and import names match, don't have. With out rechecking and reading in detail the above mentioned Packaging guide is guess the top_level.txt only needs to be present when import and package name differ or when multiple sub-packages and modules should be importable as if distinct top-level modules and packages.