perf(aggr): fuse indexed statistical groups

[singaraiona]

What changed

• extend the grouped-index slice kernel to compute variance/stddev accumulators directly
• add binary aggregate partials for Pearson correlation, covariance, weighted sum, and weighted average
• keep eligible statistical aggregate queries on the slice path instead of folding index slices back into a generic selection
• add indexed/plain and contiguous/non-contiguous parity coverage for every admitted binary aggregate

Why

Filtered group-bys over a grouped symbol index already have exact row slices for each requested key. Statistical and binary aggregate shapes were rejected by the slice kernel, forcing those slices back into a sorted selection and then through generic grouping.

Validation

• make test: 3,615/3,615 passed under ASan/UBSan
• clean release build succeeded
• 14/14 indexed/plain and sorted/unsorted statistical parity checks passed
• git diff --check passed