add atlas stream processing commands

[nickpoindexter]

Summary

Adds first-class driver support for Atlas Stream Processing (ASP), implementing the ASP driver spec. Users currently have to drop down to
Mongo::Client#database('admin').command(...) against a workspace endpoint; this PR adds a dedicated client / handles layer matching what's already shipped for PHP and Rust.

What's new

Public API (lib/mongo/stream_processing/)

• Mongo::StreamProcessing::Client — workspace-scoped client distinct from Mongo::Client. Validates the workspace URI (atlas-stream-*.<region>.a.query.mongodb.net and the .mongodb-<env>.net staging variants), enforces
  TLS, defaults auth_source to 'admin'. Static Client.workspace_uri?(uri) exposed for callers that want to gate before connecting.
• Mongo::StreamProcessing::Processors — #create(name, pipeline, **opts), #get(name), #get_info(name) for managing processors in a workspace.
• Mongo::StreamProcessing::Processor — #start(**opts), #stop, #drop, #stats(**opts), #samples(**opts) for a named processor.
• Mongo::StreamProcessing::ProcessorInfo — typed accessor over the getStreamProcessor response. Fields the spec marks as Optional (e.g. id, pipeline_version) are nullable.
• Mongo::StreamProcessing::SamplesResult — result of #samples, exposes cursor_id, documents, exhausted?.

Wire commands
All 8 ASP commands routed through Database#command against the admin database: createStreamProcessor, startStreamProcessor, stopStreamProcessor, dropStreamProcessor, getStreamProcessor, getStreamProcessorStats,
startSampleStreamProcessor, getMoreSampleStreamProcessor.

Notable spec / server alignment

• startAfter (in Processor#start options) is intentionally not serialized — the spec marks it RESERVED for future use and explicitly forbids drivers from sending it.
• Dev-server response shape deviations are accommodated, matching the workarounds in the PHP and Rust PRs (and the Python POC):
  • Processors#get_info unwraps a top-level result wrapper when the server returns { ok: 1, result: { … } }.
  • Processor#samples accepts both nextBatch (spec) and messages (current dev server).
  • ProcessorInfo#id and #pipeline_version return nil when the server omits them.
• #samples is a single-command dispatch: absent / zero :cursor_id → startSampleStreamProcessor (returns SamplesResult with cursor id, empty docs); non-zero → getMoreSampleStreamProcessor. Callers stop when
  cursor_id == 0 (or use the #exhausted? predicate).
• State strings returned as plain strings, not symbols or constants — per the spec, drivers MUST surface unknown state values as-is rather than mapping to a closed set.
• Internal-only fields (tenantID, projectId, processorId) are not surfaced in the public API.

Test plan

• bundle exec rubocop lib/mongo/stream_processing*.rb spec/mongo/stream_processing/*.rb spec/integration/stream_processing_spec.rb examples/stream_processing.rb — clean
• bundle exec rspec spec/mongo/stream_processing/ — 29/29 unit examples pass (URI detection for prod / staging / creds-and-port / case-insensitivity / non-string / four reject cases; constructor rejection paths; full
  ProcessorInfo getter coverage incl. defaults and error fields; SamplesResult predicates)
• bundle exec rspec spec/integration/stream_processing_spec.rb — self-skips cleanly without MONGODB_STREAM_PROCESSING_URI (1 pending, 0 failures)
• MONGODB_STREAM_PROCESSING_URI=… bundle exec rspec spec/integration/stream_processing_spec.rb — full lifecycle test against a real workspace endpoint. Needs an Evergreen variant configured with workspace credentials
  before it actually exercises in CI.