feat(persistence): span query DSL with SQL

[RogerHYang]

resolves #2081
resolves #2780
resolves #2782
resolves #2358

Notes

1. The UI changes are a result of resolving Consistent partitioning and nesting by the . (dot) separator for span attributes  #2081, giving us consistent JSON paths for attributes.
2. Unfortunately no. 1 has made hiding the embedding vector quite difficult. Punting for now.
3. latency_ms is changed to a calculated field in DB. Unfortunately sqlite doesn’t give the same values as python or postgres (not sure why), So sqlite values are slightly off and filtered results can be slightly different.
4. Unfortunately substring search (e.g. 'service' in output.value) in sqlite is using the LIKE operator which is case-insensitive, which gives different filtered results than python or postgres. It’s not easy to fix this. Punting for now. (Also special characters such as % are not escaped. Also punting.)
5. Postgres is strongly type. JSON values are interpreted as string by default. For custom attributes, user can now use str(), float(), int() to change the type. I have tried to infer as much as possible, but this is not always possible.
6. I’ve added a notebook under integration-tests for testing. The results match those from the main branch. One caveat is that the main branch endpoint needs to be patched from GET to POST for this to work. I’ll need to get integration tests working on CI next.
7. Also, note that JSONB doesn’t guarantee dict key ordering so the ordering will appear different than those in sqlite which stores the attributes as TEXT
8. The use_active_session_if_available option on px.Client is more difficult to now that the db is async. So i have nixed it. Now both thread and process sessions are making network calls using px.Client.
9. Also, i had to rename some of the fields in the database just to make it easier for backward compatibility. E.g. parent_span_id is renamed as parent_id
10. sqlalchemy is bounded below by 2.0.4 in order to use inplace.expression for hybrid_property.
11. openinference.span.kind is now included in the atributes. this shows the original string value in the event that it gets overwrittend to UNKNOWN because it doesn't fit the enum. openinference.span.kind is also outputted in the exported spans dataframe.

[axiomofjoy]

Really nice job, Roger. Stepping through the test cases helped me understand how the filters, query, and translator works.