Consolidate redundant Python test scaffolding

[lbx154]

Repeated Python test setup obscures the cases being checked. Consolidate the backend-resolution, model-selection and terminal-result tests into explicit parameter tables, share temporary Codex configuration setup, and use one event loop for the Grok stream. This removes a net 226 lines across six test files.

The existing inputs and runtime assertions remain covered:

Area	Retained scenarios
Binary resolution	Six backend/PATH/standard-install cases; both resolver and runner constructor checked
Pricing-model selection	Seven original precedence/empty-input combinations, now individually parameterized
Dsh completion	Successful output, empty output, nonzero exit and already-completed state
Dsh readiness	Key in the local environment file and missing credentials
Grok events	Ordered initialization/text/completion and both unsuccessful-result variants

Remove the unused journal helper and fake-process fields/methods that finalization never reads. Drop the assertion that the legacy alias package must contain exactly two source files; its module identity, monkeypatching, command-line compatibility and process-recognition tests remain. The changed files collect 101 cases versus 99 previously because the existing pricing input variations are now separate cases.

Validation: all 101 tests in the changed files passed, whole-repository Python lint passed, and diff checks passed. Existing CI workflow checks cover the full Linux Python suite and the three supported TypeScript hosts. This is a tests-only cleanup; runtime and documentation files are unchanged.