Builds fail fast on any ruby warnings

[nitishr]

Closes #729

Introduces and uses a reusable command in CircleCI config for failing fast on warnings. Addresses resulting warnings that originally caused builds to fail with the updated configuration.

Changes:

1. CircleCI Configuration:

• Introduce a reusable run-fail-fast command to execute commands and fail on any warnings in the output.
• Conditionally set RUBYOPT for all Ruby versions except ruby:2.2 (as that version doesn't support the frozen-string-literal flag).

2. Update Gemfile: Add the rdoc gem to the development group (as rdoc will no longer be part of the default gems starting from Ruby 3.5.0)
3. Fix Resulting Warnings:

• Update lib/mocha/deprecation.rb to use @logger ||= Logger.new to silence uninitialized ivar warning on older ruby versions.
• Add a JRuby-specific warning filter in lib/mocha/object_methods.rb to suppress unnecessary warnings.
• Modify test/acceptance/loose_keyword_argument_matching_test.rb to remove unused variable.

Benefits:

• Improved Maintainability: The CI configuration is more maintainable and easier to update.
• Faster Feedback Loop: The run-fail-fast command ensures that the CI pipeline fails immediately if any warnings are encountered during test execution, providing faster feedback.

Testing:

The CI pipeline has been tested to ensure that it correctly fails on warnings for the specified Ruby versions.

[floehopper]

Closes #729

Introduces and uses a reusable command in CircleCI config for failing fast on warnings. Addresses resulting warnings that originally caused builds to fail with the updated configuration.

Thank you. I'm broadly in favour of this. I've added some comments below and in some review comments.

I've fixed the formatting issues you fixed in .circleci/config.yml, so can you rebase against main to remove them from the diff?

While I understand why the commits are in the order they are, I think I'd prefer the commits with the warning fixes to come before the changes to the CI build which catches them, i.e. so all the tests pass in each commit.

Changes:

1. **CircleCI Configuration:**


* Introduce a reusable run-fail-fast command to execute commands and fail on any warnings in the output.

This seems like quite a fragile way to detect warnings. Is there definitely no better way to do this. I wonder if it's worth looking around at what other projects do? I haven't thought this through, but could we globally patch Warning#warn to raise an exception? Although I'm not sure when Warning#warn was added or whether it captures all warnings...

* Conditionally set RUBYOPT for all Ruby versions except ruby:2.2 (as that version doesn't support the frozen-string-literal flag).


2. **Update Gemfile:** Add the rdoc gem to the development group (as rdoc will no longer be part of the default gems starting from Ruby 3.5.0)

Does the absence of this gem actually trigger a warning at the moment?

3. **Fix Resulting Warnings:**


* Update lib/mocha/deprecation.rb to use @logger ||= Logger.new to silence uninitialized ivar warning on older ruby versions.

* Add a JRuby-specific warning filter in lib/mocha/object_methods.rb to suppress unnecessary warnings.

* Modify test/acceptance/loose_keyword_argument_matching_test.rb to remove unused variable.

Benefits:

* **Improved Maintainability:** The CI configuration is more maintainable and easier to update.

* **Faster Feedback Loop:** The run-fail-fast command ensures that the CI pipeline fails immediately if any warnings are encountered during test execution, providing faster feedback.

Testing:

The CI pipeline has been tested to ensure that it correctly fails on warnings for the specified Ruby versions.

Thanks for being so thorough with the testing!

[floehopper]

See review comments and comment on PR description.

[floehopper]

If it works I think it might be slightly more expressive to use > "ruby:2.2" instead of != "ruby:2.2".

[floehopper]

Unless I'm mistaken, this change is orthogonal to the other changes in this commit. If so, can you split it out into a separate explanatory commit?

[floehopper]

This looks a bit ugly. Could we avoid the public_send workaround (and associated comment) and the disabling of two cops by using an approach more like this?

As an aside, I wonder if it might separately be worth trying to extract this pattern somewhere. However, if so, I'd prefer to do that in a separate PR.