Also, set a lower minimum bound than usual, which
should hopefully help with preventing multiple copies
of `windows-sys` from sneaking into dependency trees.

PR #19 

The Unlicense/MIT syntax was originally, I believe, the only one that
was accepted. And `Unlicense OR MIT` was added later. And now apparently
tools whinge about the former. So fix it.

Closes #16

winapi has no future[1] and the rest of the ecosystem is standardizing around
`windows-sys` for lower level Windows bindings.

My main complaint had been churn, but it looks like they've slowed down.
Either way, there isn't really too much of a choice in the matter.

[1]: https://github.com/BurntSushi/winapi-util/pull/13#issuecomment-1991282893

The plan is to use this in ripgrep as part of supporting hyperlinks.

Ref https://github.com/BurntSushi/ripgrep/pull/2483#issuecomment-1724745500

Previously https://docs.rs/winapi-util went to the x86_64-unknown-linux-gnu
build of this crate by default, which is empty. This way the same link will
show useful documentation.

We achieve this by only building docs for one Windows target. Since the
API doesn't vary with other Windows targets, this seems fine.

PR #3 

For more details:
https://github.com/actions/checkout/issues/23#issuecomment-572688577

We avoid incurring a public dependency on winapi and
define our own rect struct.

PR #2 

There's not much sense for having two separate crates here. wincolor
predates winapi_util, and since wincolor is already very small (~100-200
lines), it makes sense to just roll it into here.

This adds a convenience routine for checking whether file information
has the hidden attribute set.

This was motivated by this bug report against ripgrep:
https://github.com/BurntSushi/ripgrep/issues/1154