FFE: add delivery delay test proving Agent bypass gap

[leoromanovsky]

Motivation

Datadog's feature flagging SDK (FFE) relies on Remote Configuration to deliver flag definitions to tracers. In production, tracers wait 60-130+ seconds at startup before flag configuration arrives. The root cause is a structural mismatch in the Agent: the RC cache bypass fires when a new client connects, but not when an existing client registers a new product. All tracer products (APM, ASM, FFE) share a single RC client. By the time FFE_FLAGS is registered, the client is already active -- no bypass fires, and the Agent only discovers FFE_FLAGS on its next background poll (50-60s later).

This PR adds a system-test that codifies this delay so we can prove the problem exists today, and later prove it is fixed when the Agent ships a product-aware bypass.

Changes

• containers.py: Changed the Agent RC refresh interval from a hard assignment to setdefault, so scenarios can override it via agent_env. Existing debugger scenarios are unaffected (they do not pass this key).
• __init__.py: Added a new FEATURE_FLAGGING_AND_EXPERIMENTATION_BACKEND scenario. Same as the existing FFE scenario but with rc_backend_enabled=True (configs flow through the real Agent instead of being mocked by the proxy) and a 60s Agent refresh interval to match real customer defaults. Tracer polls every 1s.
• test_dynamic_evaluation.py: Added Test_FFE_RC_Delivery_Delay. Posts an FFE_FLAGS config to the mocked backend, sleeps 5 seconds, then asserts the tracer has NOT received it. With a 60s Agent refresh and no product-aware bypass, the Agent has not polled the backend yet, so the tracer gets nothing.

Decisions

• 60s Agent refresh, not 1s: The whole point is to use the default customer configuration. With a fast refresh the test passes trivially. With 60s, the only way FFE_FLAGS can arrive in under 5s is if the bypass fires. This makes the test a proof by contradiction.
• Negative assertion (config NOT delivered): Today with a stock Agent this test should pass -- FFE_FLAGS is not delivered within 5s. When the Agent ships the product-aware bypass fix, we flip the assertion to assert delivery and the test becomes a regression gate.
• Separate scenario instead of modifying existing: The existing FEATURE_FLAGGING_AND_EXPERIMENTATION scenario uses proxy-mocked RC (no Agent in the loop). We need rc_backend_enabled=True to exercise the real Agent RC path, which requires a separate scenario.
• setdefault instead of a new constructor parameter: Minimal change. The agent_env dict is already passed through; we just needed to stop the hardcoded 5s from clobbering it.

[github-actions]

CODEOWNERS have been resolved as:

tests/ffe/test_dynamic_evaluation.py                                    @DataDog/feature-flagging-and-experimentation-sdk @DataDog/system-tests-core
utils/_context/_scenarios/__init__.py                                   @DataDog/system-tests-core
utils/_context/containers.py                                            @DataDog/system-tests-core