Use automatic batching in tabular benchmarks

[RBendias]

Remove the tabular benchmark adapter’s hard-coded estimator batching thresholds and manual batch loop. The adapter now forwards its estimator_batch_size setting directly to the model for both cached and uncached inference, so "auto" uses the library planner.

This keeps the batching infrastructure introduced in #989 while replacing the benchmark-specific 2,000/3,000-row and 50,000-cell decisions.

Depends on #1007. Related: #989 and #994.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.