At least with Tauri it's easy to both make the choice and change it later if you want to. I think the docs are sparse because it's your decision to make. I've done it both ways and there are pros and cons. If you use the sqlite plugin and write the actual statements on the JS side then you don't need to worry about the JS<->Rust interface and sharing types. Easier to just get going. If you write your own interface then you probably want to generate TS types from Rust. I think a big advantage to the Rust interface way is that it makes it easier to have the web side be dual purpose with the same code running on the web and in Tauri - the only difference being whether it invokes a tauri call or an API call.
I'll note that I have gone a slightly different path for the main app I wrote: I've written adapters on the js side that generate SQL or API calls depending on where the code is running and I wrote my own select/insert/update/delete tauri commands. The reason I ended up with what seems like a hybrid of the approaches I suggested above is that the js side knows more about what it wants and therefore generates SQL/api calls with the appropriate joins and preloads. On the tauri side I wanted to intercept at the data layer for a custom sync engine, which the frontend doesn't need to know about. However, I've ended up at that solution maybe because I added the tauri side after writing for the web.
It may be interesting with event sourcing, having the message bus + eventstore be on the rust side, and SQL projections be exposed in a sqlite db on the web side.
