Investigate dplyr rolling joins as alternative to data.table's for archive operations

[brookslogan]

If dplyr's rolling joins are performant, this may be one approach to getting away from reference semantics altogether.

[brookslogan]

Cursory initial testing revealed that dplyr took 2x the time, and polars and duckdb took similar time to, data.table, for as_of operations. polars and duckdb also involve more installation time. I still need to check if there are indexing features that might change the as_of performance comparison or extra features to consider.

[brookslogan]

We may also want to investigate

• vctrs::vec_locate_matches
• nesting by epikey, or custom vectors that allow for "fake" nesting while being primarily backed by a long vector rather than a list of vectors