New spec suite for Active Job integrations

[solnic]

Description

⚠️ This won't be done all-at-once - this issue is an anchor for all the work that must be done for the epic issue #2932. It will be growing and most likely being refined as we move forward. It will be done/closed once all the backends have been ported.

---

The existing activejob_spec.rb is a monolithic file that mixes lifecycle, context, tracing, retries and cron into one describe block and only runs against :test. It is not a suitable foundation for a rework that has to prove identical behavior across all AJ-based integrations.

The goal is to replace the original specs with a dedicated, shared-example-driven AJ spec suite under a new tree (e.g. sentry-rails/spec/active_job/). Each feature area lives in its own shared-example file. Each backend ships as a thin spec file whose entire body is roughly:

RSpec.describe "Sentry + ActiveJob on SolidQueue" do
  include_context "active_job backend harness", adapter: :solid_queue
  it_behaves_like "a Sentry-instrumented ActiveJob backend"
end

General principles and ideas:

• One top-level shared example — "a Sentry-instrumented ActiveJob backend" — composes smaller per-area shared examples. Adding a new backend is a one-line include.
• Shared examples never branch on adapter name. Per-backend quirks go through a harness helper (boot, enqueue, drain, reset) that each backend's spec file supplies via let or shared context.
• Backend-specific behaviors (e.g. sidekiq retry middleware, solid_queue recurring jobs) live in the respective backend-specific dirs or dedicated gems (sentry-sidekiq).

Feature areas the shared suite must cover

• Harness & scaffolding. New spec tree, shared context for adapter setup / teardown / draining, fixture jobs (normal, failing, retryable, discardable, scheduled, mailer, rescue-from, deserialization-error, nested-enqueue), transport/event helpers. No behavioral assertions yet — just the rails the rest of Phase 0 runs on.
• Job lifecycle. perform_now, perform_later + drain, set(wait:) / set(wait_until:) scheduled jobs, retry_on, discard_on, rescue_from (both suppressing and re-raising), ActiveJob::DeserializationError root-cause unwrapping, ActionMailer::MailDeliveryJob path.
• Error capture & context. Exception type/message/stacktrace round-trip, extra payload (active_job, arguments, scheduled_at, job_id, provider_job_id, locale, queue, executions), tags (job_id, provider_job_id, queue), scope cleared between jobs, rescue_from-suppressed errors produce no event, errors raised from rescue_from callbacks produce both events.
• Consumer-side tracing. Transaction name = job class, source: :task, op + origin conventions, status: ok / internal_error, event linked to transaction via same trace_id, messaging span data (messaging.destination.name, messaging.message.id, retry count, latency, provider job id), gated correctly by traces_sample_rate.
• Producer-side tracing. Enqueue within an active transaction produces a queue.publish child span; span carries the same messaging data; scheduled enqueue does the same. (Pending until Phase 1.)
• Distributed tracing. Producer trace headers round-trip through the payload; consumer transaction's trace_id matches producer's; consumer's parent_span_id points at the producer's publish span; baggage survives. Verified per backend — this is the shared suite's most important test for adapter conformance.
• Scope & context propagation. Hub cloned per job, user set on producer scope surfaces on consumer scope, tags propagate, no scope leak between consecutive jobs, no leak across threads/fibers.
• Retry & discard semantics. active_job_report_on_retry_error true/false, final-attempt-only reporting, discarded jobs emit no event, retry count reflected in transaction span data and tags, exhausted-retry capture.
• Argument serialization. Ports the existing GlobalID / nested / Range / TimeWithZone / raising-to_global_id coverage from the old file into a dedicated shared-examples file. Adds sensitive-value handling (send_default_pii) if applicable.
• Concurrency. Parallel job execution keeps scopes isolated; thread-local hub is not shared; trace ids do not cross-contaminate.
• Backend wiring. One thin spec file per adapter: :test, :inline, :async, solid_queue, good_job, sidekiq. Each just configures the adapter via the harness and includes the top-level shared example.

[linear-code]

RUBY-171