Fix compatibility issues with various versions of Rust.

[Anders429]

Fix for #5. This allows the crate to be used in all versions of Rust, from 1.0.0 upwards, as specified by the Rust version badge. I have also added coverage for each of these versions to the CI, so that they don't accidentally break again in the future :)

I had to remove the Kleene operator (?) from assert_matches! in order to not lose coverage in rustc 1.32.0, since the operator requires edition 2018 in order to be used before rustc 1.37.0. I simply broke the optional arguments into separate matchers. Interestingly, this allows assert_matches! to be used in earlier rustc versions. I bumped the MSRV for it back to 1.26.0, because that's as far back as it tests, but you could probably move it even further back by making the tests compatible (it builds all the way back to version 1.7.0, and the only thing holding it back is that the tests aren't compatible before 1.26.0). For now, I've left it at 1.26.0.

[Anders429]

Hmm, I think the no_std failure is a shortcoming of cargo-nono. Building the crate on a no_std target works just fine:

$ cargo build --target thumbv6m-none-eabi
    Finished dev [unoptimized + debuginfo] target(s) in 0.01s

Seems like cargo-nono doesn't check for anything besides #![no_std] or something like #![cfg_attr(not(feature = "std"), no_std)]. Going off of this issue, it doesn't seem like other cases are covered in its checks.

[Anders429]

After further investigation, the breakage is indeed due to a limitation of cargo-nono. The code only checks for #![no_std] and #![cfg_attr(not(test), no_std)] explicitly, and doesn't check for any other cfg_attr form. There is one other check for features that disable no_std, but that is all.

[Anders429]

I've changed the no_std test in the CI to directly build the crate on a no_std target, thumbv6m-none-eabi. This is, in my opinion, better than running cargo-nono, as it actually runs the build and will fail if there are any actual errors in building on no_std.

This is the same approach used by many other crates, including rand.

If you'd rather not do this, that's fine with me, but this resolves the CI failure while still showing no_std compatibility :)

[svartalf]

If you'd rather not do this, that's fine with me

Also fine by me, thanks for doing it!
Can you elaborate on one thing I personally find confusing?

[svartalf]

I'll publish new version in a couple hours, thanks for your work!

[svartalf]

Follow-up: your changes were released in claim = "0.5.0" release