Improving the Benchmark Suite

The benchmark suite is a highly flexible performance benchmarking framework, centered
around Shake build targets. However, it has high cognitive load for HLS non-experts. In
particular, it:

* Has multiple interconnected projects: `bench`, `ghcide-bench` and `shake-bench`.
* Constructs an entire Shake build system
* Outputs large hierarchical CSV
* Has high levels of abstraction

That said, I propose the following changes:

* Consolidate `bench`, `ghcide-bench` and `shake-bench` into a single `bench` project
* Completely remove `shake`
* Replace orchestration/statistical processing with `tasty-bench`, if reasonably fits
* Use familiar benchmarking terminology:
* Examples -> Projects or Harnesses
* Experiments -> Benchmarks
* Configurations -> separate into two distinct concepts:
* Test run configuration
* HLS plugin set

Start by mapping the three named projects—bench, ghcide-bench, and shake-bench—and reviewing how they construct Shake builds and produce hierarchical CSV output. Compare the current orchestration with tasty-bench, then define completion as one consolidated bench project with the proposed terminology and no Shake dependency.

Written by the indexing model from the issue text.