Refactor start and end range regression check fails to Error from panic

[chrissimpkins]

This PR refactors the panics on start and end date regression checks in the least_satisfying function to failure::Error and adds error handling so that this propagates to the run() function.

The user receives the error message and non-zero exit status code rather than a panic:

From:

verifying the end of the range reproduces the regression
std for x86_64-apple-darwin: 16.39 MB / 16.39 MB [============================] 100.00 % 11.44 MB/s uninstalling nightly-2020-02-03
tested nightly-2020-02-03, got No
thread 'main' panicked at 'the end of the range to test must reproduce the regression', /Users/chris/.cargo/registry/src/github.com-1ecc6299db9ec823/cargo-bisect-rustc-0.2.1/src/least_satisfying.rs:34:14
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace.

to:

verifying the end of the range reproduces the regression
std for x86_64-apple-darwin: 16.39 MB / 16.39 MB [============================] 100.00 % 13.99 MB/s uninstalling nightly-2020-02-03
tested nightly-2020-02-03, got No
Error: the end of the range to test must reproduce the regression

[chrissimpkins]

Closing in favor of #58 that includes broad support for error handling and implements the changes here.