CRTP matchers

[dvirtz]

Description

As discussed in #1553, the current matchers implementation forces passing values by const ref. Changing the implementation to be CRTP based instead of dynamic polymorphism, enables forwarding the checked value as it is.
A nice benefit is that it is no longer required to explicitly set the value type for the Predicate matcher.

GitHub Issues

#1553

[codecov]

Codecov Report

Merging #1554 into master will decrease coverage by 0.2%.
The diff coverage is 30%.

@@            Coverage Diff            @@
##           master    #1554     +/-   ##
=========================================
- Coverage   80.54%   80.33%   -0.2%     
=========================================
  Files         121      118      -3     
  Lines        3386     3361     -25     
=========================================
- Hits         2727     2700     -27     
- Misses        659      661      +2

[codecov]

Codecov Report

Merging #1554 into dev-v3 will increase coverage by 5.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           dev-v3    #1554      +/-   ##
==========================================
+ Coverage   89.28%   94.29%   +5.01%     
==========================================
  Files         139        8     -131     
  Lines        5997       70    -5927     
==========================================
- Hits         5354       66    -5288     
+ Misses        643        4     -639

[dvirtz]

I've renamed the new generic matcher to MatcherBaseGeneric and restored MatcherBase for keeping backward compatibility.
I also added more tests taken from #1843 (thanks @melak47).
All the custom matchers are implemented with MatcherBaseGeneric which should give them better performance as no virtual functions are used.

[melak47]

In my PR I found I could get away without CRTP, just having a base class from the namespace where the non-member operator overloads are defined is enough to get them picked up via ADL.

(With a non-CRTP MatcherBase I think you could replace the existing virtual machinery with very little to no breaking changes. Only users who explicitly used override in their custom matchers would get an error. I don't know if this would be an acceptable amount of breakage)