Potential improvements for _SearchIndexer

[vyasr]

683 replaced the old Collection class with a stripped down _SearchIndexer that does just enough for signac's internal use cases. However, that PR left a few tasks outstanding that we should consider:

• [ ] Replacing the _SearchIndexer with a plain dict and free functions. I don't anticipate that these changes will have a significant effect on the code base, but it's worth considering since it may improve clarity. The _SearchIndexer has essentially no state beyond that of a normal dict, and while its contents are more restricted there is no particular enforcement, so conceptually using a free function might be clearer. Additionally, it is possible to create the object in an invalid state with the current approach. Some additional discussion is here.
• [ ] Remove the potentially unnecessary JSON normalization as discussed in https://github.com/glotzerlab/signac/pull/667#discussion_r805197820 (currently found here). I believe that this function is only invoked in places that have already performed suitable normalization, but we should verify this before making any final changes.
• [ ] Renaming the _SearchIndexer. This is the lowest priority task and would be moot if we reimplemented it as a dict with free functions.

[bdice]

These are good ideas. Since all this is internal, this work can be done post-2.0.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[vyasr]

FWIW just tried running the test suite with the normalization removed, and everything passed. I also tried running it throwing an error whenever the normalization changed the input, and the only difference I observed was a conversion from tuples to lists. In terms of performance, the normalization costs a few microseconds for a trivial filter (e.g.{'a': [1]}) rising to 10s of microseconds for filters including more inputs.