bench(pingora-lru): migrate benchmarks to Criterion

[C-Harshul]

Closes #602

The existing benchmarks in pingora-lru use manual Instant::now() timing inside a fn main(). This produces a single number with no warmup, no outlier handling, and no way to detect regressions automatically. This PR migrates both benchmarks to Criterion.rs.

Changes

pingora-lru/Cargo.toml
Adds criterion = { version = "0.5", features = ["html_reports"] } to [dev-dependencies].

benches/bench_linked_list.rs
Migrates all five operations — push, iter, search, promote, and pop — into named Criterion benchmark groups. iter_batched(LargeInput) is used for the pop benchmark so the cost of building the list is excluded from the measured time.

benches/bench_lru.rs
Migrates single-thread and 8-thread concurrent benchmarks. iter_custom is used for the concurrent case since Criterion does not manage threads natively — it times the wall-clock duration of all threads completing iters operations each, which mirrors the intent of the original benchmark.

Before / after

	Before	After
Output	8ns avg per operation	[8.1 ns 8.3 ns 8.6 ns]
Warmup	✗	✓
Confidence interval	✗	✓ 95%
Regression detection	✗	✓ automatic
Outlier handling	✗	✓