temporal: Emit accurate Ruby worker slot metrics

[kstewartfigma]

Why

Ruby Temporal worker processes currently emit thread_pool.available_threads into the same DogStatsD metric context. With multiple worker processes sharing a pod or host, the last gauge write wins, so the metric does not represent fleet slot saturation.

What changed

• preserve the existing thread_pool.available_threads metric for compatibility
• add Temporal-standard worker_task_slots_available and worker_task_slots_used gauges
• tag the new gauges with namespace, task_queue, worker_type, and a unique worker_id so concurrent processes cannot overwrite one another
• report an initial snapshot and refresh it every 10 seconds, including while idle
• stop the reporter before enqueueing shutdown sentinels so shutdown does not appear as real slot use

Test plan

• bundle exec rspec spec/unit/lib/temporal/thread_pool_spec.rb spec/unit/lib/temporal/activity/poller_spec.rb spec/unit/lib/temporal/workflow/poller_spec.rb
• 28 examples, 0 failures
• ruby -c on all changed Ruby files

The full suite has 10 unrelated failures under local Ruby 3.4 that reproduce on an untouched master worktree. This repository's GitHub workflow runs Ruby 3.0.3.