Make useNamingConvention's regexes non-greedy by default

[Conaclos]

useNamingConvention introduced a way to write custom conventions.
A user can enforce any style thanks to regexes. The provided regexes are said restricted because we carefully expose the features we want.
A restricted regex is implicitly bounded by the ^ and $ anchors.
To make thing clear, I will include these implicit anchors in the following examples.

For now, we support both greedy and non-greedy quantifiers.
In most cases, the choice between greedy and non-greedy quantifiers doesn't change the meaning of a regex because of the implicit anchors.
For example, ^My(.+)Error$ and ^My(.+?)Error$ exhibit the same behavior.

Unfortunately, in some cases, behavior doesn't match. For example ^(.+)x?$ and ^(.+?)x?$ doesn't exhibit the same behavior.
If OI take the string testx, the first regex captures the entire string, while the second one only captures test without x.

I think that users always expect non-greediness. Otherwise, this makes $? completely useless.

Thus, I think we should default to non-greedy quantifiers: .* could be non-greedy, instead to be greedy.

Unfortunately, the regex crate only allows you to swap the greediness behavior using swap_greed. This means that .* becomes non-greedy, while .*? becomes greedy.
This is a breaking change from a user POV.

Another solution is to rely directly on the regex-automata crate (that is a dependency of the regex crate). I have mixed feelings about this option because the crate is still unstable - even if we can expect the crate to be stable enough. regex-automata allows to use a non-greedy matching strategy using Inout::earliest.

Once, we make the non-greedy behavior the default, I think we could deprecate the quantifiers followed by a question mark (.*?, .+?, ...).

[Conaclos]

Note: We might consider making the breaking change a minor one, since non-greedy quantifiers are not likely to be used as much.