Reuse forward histograms in Triton TBE backward (#4491)

Summary:
Pull Request resolved: #4491

Reuse exact counts from an eligible leading, unique small-table feature during exact rowwise Adagrad backward. Immutable plans carry actual table geometry, compensated FP16 high/residual matrix products preserve FP32 accuracy, and scalar validity enables safe fallback for long bags, aliases, graph capture, weighted inputs, and VBE.

Reviewed By: axeisghost

Differential Revision: D114284800

fbshipit-source-id: b7d9f3fbb962bf3e8e25b5bc46f32024e43c0ed8

Author: stashuk-olek