feat(embed): add query runner

[oliverb123]

First step to accessing the general embeddings table from hogql queries, which unlocks fuzzy search and similarity work across products. There are still details to be worked out:

• Exposing product, document_type, document_id, timestamp and distance for user-facing/dynamic use in WHERE and ORDER BY clauses. I think think means a new category of taxonomic property filter, and I simply didn't want to take on that headache right now, so there's a hacky half solution in place now. Callers from inside the django app can of course add whatever filters they like right to the AST prior to joining/calculating.
• Add a k8s service to the embedding worker deployment, so talking to it works in prod

Also it lets users use us as a vector DB if we want (we'd need to give them a way to write to it tho):

ok maybe we're cooking

[github-actions]

Size Change: +53 B (0%)

Total Size: 3.05 MB

ℹ️ View Unchanged

Filename	Size	Change
frontend/dist/toolbar.js	3.05 MB	+53 B (0%)

_{compressed-size-action}

[greptile-apps]

_{9 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[daibhin]

Left a couple of comments. I'll review the query runner properly on Monday. Might be worth getting the DW folks to look at the HogQL changes

[Gilbert09]

Just update the table name - the rest of the hogql looks good!

[daibhin]

Left a few comments, mostly around explaining how the query works given it's relative complexity. The API is pretty intuitive but the query itself is harder to understand. Maybe tests would be the best description