When routing a request, `#find_routes` will iterate through potential
routes for the request and yield them to the passed block. `#serve` and
`#recognize` both used `#find_routes`, but ended up handling things
slightly differently. Both methods would modify the `script_name` and
`path_info` of the request if the current route is not anchored.
However, only `#serve` used `#chomp` and would restore the request's
original `script_name` and `path_info` after each iteration. This
inconsistency can cause issues as `#recognize` can end up not matching
them even though `#serve` would.

This commit fixes the issue by unifying more of the implementations of
`#serve` and `#recognize`. Now, `#find_routes` implements the
mutation/restoration of request params for both `#serve` and
`#recognize` so that their behavior is consistent and request mutation
isn't leaked between iterations.

One test had to be changed because it depended on the request mutation
leaking outside of `#recognize`, so it was updated to make assertions
during `#recognize` instead.