With the changes we've made recently, the fuzzer has been running out of
memory by discovering very large padding 😂

In this commit, I attempted to fix these fuzzing failures by using the
`strftime::io` and `strftime::fmt` variants. In `strftime::fmt`, the
fuzzer immediately found a crash.

If a multibyte UTF-8 sequence appears immediately following a `%` spec
begin marker, the loop to pluck out padding consumes one element from
the cursor and then breaks. This leads to pushing a partial UTF-8
character into the `fmt::Write` instance, which causes a panic in the
underlying `String` because the string's contents is now malformed
UTF-8.

Add a test and break with `None` if encountering a non-ASCII byte while
processing the spec.

I've also cleaned up the fuzz targets so they build without warnings.