Update pyproject.toml for release on PyPI

[khl02007]

Tried publishing on test-pypi and seems to work

[rly]

Note that this also upgrades the hdmf version to the latest (>=3.4.2) which includes a fix for the bug reported in #260 and hdmf-dev/hdmf#739

This also uses the stable published release of spikeinterface instead of the latest development branch because we are not allowed to publish a python package that has an unreleased dependency.

[edeno]

My main concern is the change to the latest published version of spikeinterface. As I recall, we changed to installing from the spikeinterface repo to quickly take advantage of new features we requested. Also, I believe the release schedule for spikeinterface has been slow relative to their features. So if we revert back to the latest published version, this may break parts of the pipeline. One possible solution is to just have users install from the repo after installing from pip or conda, but this isn't ideal. @khl02007 and @rly , do you have thoughts on this?

[rly]

I agree. I think we should

1. request spikeinterface make more frequent releases
2. include in the installation instructions that users should install from the spikeinterface git repo after installing spyglass, and pin to a particular commit that we know is working. This is not ideal, but is the only way to use unreleased features from spikeinterface. Alternatively, instead of pinning to a particular commit, we could let pip install the latest main branch, but then an uncaught bug in the main branch may affect our users and no one would know. IMO it is slightly better to recommend installing from a commit that we know is working. Creating identical environments for debugging also becomes easier. A major disadvantage is that we would have to update that commit hash in the instructions periodically to the latest commit, but I think that is OK.