RUST-1604 Add custom bucketing fields to timeseries options

[isabelatkinson]

The drivers ticket description only details a test sync, but the new tests were using options we hadn't implemented yet.

[abr-egn]

Is this a backwards-compatible change?

[isabelatkinson]

Good question. I did some testing and it looks like setter(into) will break this:

let options = TimeseriesOptions::builder()
        .time_field("x".into())
        .meta_field(Some("y".into()))
        .granularity(Some(TimeseriesGranularity::Hours))
        .build();

so we'll need to leave that out for now. We could also override the default setting for the existing fields using #[builder(setter(!into))] (see docs), but the inconsistency between fields might be confusing.

The lack of default here, however, is currently pretty prohibitive, as the following won't compile:

let options = TimeseriesOptions::builder().time_field("x".into()).build();

because the builder doesn't default the optional fields to None. This means that users currently need to specify all fields, and therefore adding new fields will break code. I can't think of a scenario in which default will break anything so I think it's a fine addition, but if we want to play it safe we could just defer these fields until 3.0.

As an aside, I know we decided a while back that these we should use default and setter(into) for all of our TypedBuilder implementations, but I can't find any documentation of this choice outside of (#368). I remember some convos about this in the team channel and the Prisma channel, but those would've been cleared a long time ago because of the slack retention policy. It looks like there are some structs in our API that don't have these options set, so I filed RUST-1695 to fix this for 3.0 and document/enforce the policy to avoid running into this in the future.

[abr-egn]

Thanks for testing this! I also can't see any way default would be breaking, so that seems like the right way here.

[abr-egn]

LGTM!