Benchmark improvements

* CI: Clean offline; display cleaned bytes; one artifact upload
* Add missing csv bench to iai benchmarks
* Generate flamegraphs for benchmarks
* Validate benchmark results separately

Cargo.toml

@@ -115,9 +115,20 @@ serial_test = "2.0.0"
 iai-callgrind = "0.7.3"
 criterion = "0.5.1"
 
+[target.'cfg(not(target_os = "windows"))'.dev-dependencies]
+pprof = { version = "0.13.0", features = ["criterion", "flamegraph"] }
+
 [patch.crates-io]
 modular-bitfield = { git = "https://github.com/mthom/modular-bitfield" }
 
+[profile.bench]
+lto = true
+opt-level = 3
+
+[profile.release]
+lto = true
+opt-level = 3
+
 [[bench]]
 name = "run_criterion"
 harness = false