Clear checkout state when shutdown cleanup raises

[OskarEichler]

Summary

Always clear the final checked-out connection reference and nesting count, including when a shutdown cleanup callback raises during checkin.

The callback exception still propagates. The change prevents a subsequent checkout from bypassing the closed pool and returning the stale connection left in the caller's local state.

Reproduction

require 'connection_pool'
pool = ConnectionPool.new(size: 1) { Object.new }
pool.checkout
pool.shutdown { raise 'close failed' }
begin
  pool.checkin
rescue RuntimeError
end
pool.checkout

Before: the last call returns the old connection, despite shutdown.
After: it raises ConnectionPool::PoolShuttingDownError.

Verification

• Ruby 4.0.6: existing suite passes with the release-based patch, 86 runs / 187 assertions / zero failures or errors.
• Focused ordinary and forced final-checkin cases confirm cleanup error propagation and rejection of subsequent checkout.
• Existing targeted StandardRB, Ruby syntax and git whitespace checks pass.
• This PR contains only the cleanup correction on main. The consumer patch is based on installed v3.0.2.

Limitations and breaking changes

• Used an external verification Gemfile to add Minitest 6's extracted minitest-mock support (also present in current upstream). The unmodified release setup initially had four missing-stub errors. No test/Gemfile/gemspec changes included.
• No new tests under the consumer's no-new-tests policy; existing tests and the focused reproduction were run.
• No JRuby/TruffleRuby/Linux runtime checks were run locally.
• Breaking changes: none intended. Shutdown callback exceptions still propagate; subsequent checkout now honors shutdown. Nested checkin and discarded-connection cleanup retain their existing behavior.

Prepared with Codex assistance.