refactor(tests): more test stability fixes and improvements

[solnic]

I had to make more tweaks for simplicity and better stability. There are cases where it simply must not be allowed to use async: true as tests are relying on changing configuration of globals that we don't own and control.

It's worth mentioning that our end-goal here should be to introduce a proper client/transport split like we have in other SDKs. This would be a cleaner architecture with easier testability being a nice side-effect. Leaving this for 14.0.

This PR can be reviewed commit-by-commit.

Summary

• Rework Sentry.Test isolation around an explicit Scope abstraction. Replace the flat :persistent_term layout (one key per {pid, config_key} plus separate scope-marker, allow-map, and counter keys) with a Sentry.Test.Scope value struct (one entry per test) and a Sentry.Test.Scope.Registry that owns lifecycle and resolution. Sentry.Test.Config becomes a thin adapter — namespace/1, put/1, allow/2 keep their signatures.
• Drop the single-active-scope fallback. Master's resolve_from_active_scopes/1 routed any unclaimed pid to whichever test happened to be active, which caused stray events (e.g. OTP's async callback-crashed meta-event) to leak across tests. The new resolution tries three precise strategies and stops: walk [pid | $callers], walk $ancestors against each scope's owner_pid, walk [pid | $ancestors] against each scope's allowed_pids. Unmatched pids fall through to :default.
• Transparent routing for globally-supervised pids. Sentry.Test.Config.put/1 auto-soft-allows :logger, :logger_sup, and Sentry.Supervisor onto the calling scope so downstream suites get per-test routing with zero opt-in. Soft-allow silently skips when a live peer scope already owns a global (async-safe); explicit allow/2 keeps strict conflict detection via Scope.AllowConflictError.
• O(1) resolution via a reverse-index :persistent_term key. Lookups no longer iterate active scopes; lookup_allow_owner/1 and fetch/1 are direct gets. This keeps Sentry config reads cheap under async load.
• Ancestor-lookup improvements. Warn on ambiguous overrides, dedupe collect_ancestors/3 across siblings, tighten ancestors_of/1, gate the OTP 25+ parent-pid probe behind a tag.
• Longer default receive_timeout in Sentry.Case so async suites don't flake under load.
• :enable_logs handler-config hardening. Reject non-boolean overrides at attach time; add positive-path coverage for handler-level enable_logs: false overriding a global enable_logs: true.
• DSN handling in Sentry.Test.Registry. Make the Bypass DSN override unconditional so a leaked SENTRY_DSN can't ship synthetic events to a real endpoint; warn when overriding; harden with a catch-all and regression tests.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 99019f2. Configure here.}