Report will_be_retried from the retry filter's decision

[Enceradeira]

Description

Fixes #1905. Companion to cucumber/cucumber-ruby-core#345, and an alternative to #1907 / cucumber/cucumber-ruby-core#344.

With --retry 1 --retry-total 2 and three failing scenarios, the third scenario is missing from the HTML report. The TestCaseFinished message of that scenario claims willBeRetried: true, because the flag was computed from --retry alone, but the retry filter, honouring --retry-total, never retries it. The HTML formatter hides an attempt that will be retried and waits for one that never comes.

On main the messages are broken in two further ways, which #1907 does not address:

• The runner is given the number of retries where it expects the number of attempts. With --retry 1 willBeRetried is never true, so a retried attempt is reported as a final result and the report shows the scenario twice.
• The first attempt's TestCaseFinished is emitted after the retry, with the started id of the retry (fixed in the core PR).

This PR introduces Cucumber::RetryPolicy, shared by the retry filter and the runner. The filter asks it whether a failed test case is retried and records what it did, the runner asks the same question when it reports the outcome. will_be_retried? has no side effects, so both get the same answer.

A new scenario in features/docs/cli/retry_failing_tests.feature asserts the message values (attempt, willBeRetried, and that every testCaseFinished references the started id of the same attempt), which no existing test did.

Notes

• The Gemfile temporarily takes cucumber-core from the branch of Let a retry policy decide will_be_retried in the runner cucumber-ruby-core#345 so CI can run. To be removed once core 20 is released, together with raising the upper bound in cucumber.gemspec.
• attempt is emitted 1-based while the messages spec says the first attempt should be 0. Pre-existing and out of scope here.

Type of change

• Bug fix

Checklist:

• Tests have been added for any changes to behaviour of the code
• New and existing tests are passing locally and on CI
• bundle exec rubocop reports no offenses
• RDoc comments have been updated
• CHANGELOG.md has been updated