Fix path handling for non-conforming datasets

[GreenK173]

This PR replaces PR #79 because I accidentally created that one on the main branch of my fork.

Fixes #78. The matter was also recently dealt with in the SigMF specification repo in issue #321 and PR #322. In particular this PR addresses the issue raised by @lgoix in this comment.

The get_dataset_filename_from_metadata function returns the full file path for non-conforming datasets, not just the file name. Furthermore an unit test verifying this, plus a couple of typos described in the above issue fixed.

[Teque5]

While rewriting your test I observed that to use a non-compliant data file we have to do something like this:

meta = SigMFFile(metadata=ncd_metadata, data_file=data_path)
# tell SigMF that the data is noncompliant
meta.set_global_field(SigMFFile.DATASET_KEY, data_path.name)

but really I think the SigMFFile should detect that data_path.suffix != '.sigmf-data' and set that key & value automatically.

We can either merge this and create another PR, or if you want you can add this improvement to this PR.

[Teque5]

Since we never want temp files to exist after testing I used tempfile and also make some simplifications.

[GreenK173]

Thanks for all the feedback.

I've implemented the automatic detection of NCD discussed above, and also made a small fix in get_dataset_filename_from_metadata.

However there's another issue: the behavior of os.remove differs between Windows and Unix platforms (see the linked doc). I'm guessing that you're using Linux where it should work fine, however on Windows (which I'm using) it crashes because the dataset file is still being used by self._memmap of both SigMFFile instances and the deletion isn't allowed. Since there seams to be no good way how to close np.memmap (see numpy/numpy#13510 for more details), the only way how I was able to solve it is to get rid of both SigMFFile instances so that they get garbage collected.

Since I've also noticed another Windows/Linux problem elsewere, I've opened issue #89 regarding this.

[Teque5]

I don't love path handling like this, but I opened #90 and we can address it in the future.

[Teque5]

Thank you @GreenK173 for your work on NCD handling and this PR.