Add medium_cells benchmark

This patch adds a new benchmark which aims at filling a gap between the
`dense_cells` and the `light_cells` benchmarks, providing a more
realistic example of escape sequence density in a normal terminal
workflow.

The new benchmark is an Alacritty ref test recording of a short NeoVim
session, which includes several small and large buffer movements,
different styles of underlines, and window-in-window using a fuzzy
finder and tiled buffers.

This new test intentionally uses a workflow that is pretty escape
sequence heavy, while still being realistic, since workflows without
heavy escape sequence use are already reasonably well covered by the
`light_cells` benchmark.

Closes #40.

benchmarks/medium_cells/benchmark

@@ -0,0 +1,5 @@
+#!/bin/sh
+
+# Write an escape-heavy vim ref test session.
+
+cat "$(dirname $0)/vim_session"

src/bench.rs

@@ -154,7 +154,7 @@ impl Benchmark {
 
         let mut samples = Vec::new();
 
-        let max_samples = max_samples.unwrap_or(usize::max_value());
+        let max_samples = max_samples.unwrap_or(usize::MAX);
         let end = Instant::now() + Duration::from_secs(max_secs);
         for _ in (0..max_samples).take_while(|_| Instant::now() < end) {
             let duration = self.run_sample(&mut stdout);
@@ -263,7 +263,7 @@ impl Results {
             return 0.;
         }
 
-        let mean = self.mean() as f64;
+        let mean = self.mean();
         let len = self.samples.len();
         self.samples.iter().map(|&s| f64::powi(s as f64 - mean, 2)).sum::<f64>() / (len - 1) as f64
     }