Improve benchmark progress output formatting

[dantegd]

This PR improves benchmark console output so long benchmark runs are easier to follow. Instead of printing repeated per-run DataFrames and noisy progress blocks, the benchmark runner now emits a concise, aligned progress table with one row per logical benchmark variation.

The new output shows the algorithm, workload shape, estimated input size, backend timing columns, and compact details such as accuracy or speedup. Mixed backend runs are grouped onto a single row when possible, making CPU/GPU comparisons much easier to scan.

Builds on top of #7980

Example output

Benchmark mode: GPU mode (cuML available)
Using CUDA Memory Resource...
 progress  algorithm                            shape        data   gpu_time   cpu_time  accel_time  details
------------------------------------------------------------------------------------------------------------
    [1/3]  RandomForestClassifier           3.3M x 16    ~0.21 GB      1.30s          -           -  acc=0.9948
    [2/3]  RandomForestClassifier        800.0K x 128    ~0.41 GB     0.370s          -           -  acc=0.9859
    [3/3]  RandomForestClassifier        200.0K x 512    ~0.41 GB     0.145s          -           -  acc=0.9472

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.