Run profiling under perf evals times

I don't know how accurate perf is for functions that take nanoseconds. My understanding is that running a function `evals` times should take at least a couple microseconds so I think this should improve profiling accuracy with neglibile extra time taken.

docs/src/manual.md

@@ -85,7 +85,7 @@ You can pass the following keyword arguments to `@benchmark`, `@benchmarkable`,
 - `gcsample`: If `true`, run `gc()` before each sample. Defaults to `BenchmarkTools.DEFAULT_PARAMETERS.gcsample = false`.
 - `time_tolerance`: The noise tolerance for the benchmark's time estimate, as a percentage. This is utilized after benchmark execution, when analyzing results. Defaults to `BenchmarkTools.DEFAULT_PARAMETERS.time_tolerance = 0.05`.
 - `memory_tolerance`: The noise tolerance for the benchmark's memory estimate, as a percentage. This is utilized after benchmark execution, when analyzing results. Defaults to `BenchmarkTools.DEFAULT_PARAMETERS.memory_tolerance = 0.01`.
-- `enable_linux_perf`: If `true`, profile using perf once. Defaults to `BenchmarkTools.DEFAULT_PARAMETERS.enable_linux_perf = false`.
+- `enable_linux_perf`: If `true`, profile using perf `evals` times. Defaults to `BenchmarkTools.DEFAULT_PARAMETERS.enable_linux_perf = false`.
 
 To change the default values of the above fields, one can mutate the fields of `BenchmarkTools.DEFAULT_PARAMETERS`, for example:
 

src/execution.jl

@@ -604,8 +604,11 @@ function generate_benchmark_definition(
                 $(setup)
                 try
                     $LinuxPerf.enable!(__linux_perf_bench)
-                    # We'll just run it one time.
+                    # We'll run it evals times.
                     __return_val_2 = $(invocation)
+                    for __iter in 2:__evals
+                        $(invocation)
+                    end
                     $LinuxPerf.disable!(__linux_perf_bench)
                     # trick the compiler not to eliminate the code
                     if rand() < 0