bench: fix broken benchmark

I'm not sure how this ever worked? Apparently, a mutable ref to `corpus`
was being taken and thus emptied out after the first use. We fix it by
re-assigning for each run.

Ref #179

bench/src/bench.rs

@@ -244,9 +244,10 @@ fn sentences(c: &mut Criterion) {
 fn byte_lines(c: &mut Criterion) {
     use bstr::io::BufReadExt;
 
-    let mut corpus = SUBTITLE_EN_HUGE;
+    let corpus = SUBTITLE_EN_HUGE;
     define(c, "bstr/for_byte_line", "ascii", corpus, move |b| {
         b.iter(|| {
+            let mut corpus = corpus;
             let mut count = 0;
             corpus
                 .for_byte_line(|line| {