Benchmark `epix_slide` for lag extraction

[brookslogan]

For backfill-aware modeling and revision analysis, we often want to extract data for a time_value of ref_time_value - lag, as_of version ref_time_value, for all ref_time_value in requested ref_time_values and all lag in requested lags. We can do this with epix_slide(), but its performance may be too slow for this purpose (vs. pseudoprospective forecasting). We should benchmark it for this purpose and judge whether its performance is acceptable.

• If Consider dedicated archive function for extracting lags #375 is complete and uses a custom implementation (not epix_slide()) with acceptable performance, then no need to do this analysis; just close as not planned.
• Perform benchmarking.
• Contextualize benchmarking for different uses: backfill-aware modeling in exploration, backfill-aware modeling in production, revision analysis (summary statistics). Is it too slow to be usable? Is it usable, but slow enough to be annoying? Or is its performance not a notable issue?
• If Consider dedicated archive function for extracting lags #375 is incomplete, update its priority based on findings. If it is complete and is based on epix_slide() and is too slow, open another issue to fix it.