Allow users to request repetition index be cached in 2.1

[westonpace]

Currently we cache the repetition index for small data types (mini block layout) and pay the 2-IOP cost for large data types (full zip). If the data is in remote storage and users have enough RAM then it may make sense to cache the repetition index in all scenarios.

[Xuanwo]

#4104 is implemented so that all repetition indexes are cached in memory when available. Do you think we should add a configuration option for this?

[westonpace]

Yes, I think we can expose it in https://github.com/lancedb/lance/blob/baa89013fd5b580ac9ffeaff339fb1554e6957c4/rust/lance-file/src/v2/reader.rs#L315

[westonpace]

Now that #4287 has merged is this ready to be closed?

[Xuanwo]

Now that #4287 has merged is this ready to be closed?

Yeah, I'll expose this setting to python and see how it works.