Isolate shotovers tokio runtime in integration tests

[rukai]

Makes integration tests more robust by isolating:

• Shotover async runtime from test runtime
• Shotover async runtime from other shotover runtimes created at the same time or afterwards

It also simplifies our test helper logic.

An immediate practical benefit is removing the ERROR shotover_proxy::observability: Metrics HTTP server failed: Failed to bind to 0.0.0.0:9001 error log clutter that occurs every time we create a second Shotover instance after destroying the previous shotover instance.

[rukai]

Oooh the redis_int_tests::cluster_dr failure actually reveals a bug in shotover that was hidden before.

[rukai]

These are fixes needed to land this:

• pass flush flags to inner chains - Pass flush flag through sub chains #937 - needed to flush the coalesce transform
• fix connection shutdown - Fix connection shutdown logic #933 - needed to keep the tokio runtime going until the connections have shutdown

[rukai]

Although the plan is to eventually remove ShotoverManager, replacing it with ShotoverProcess, it makes sense to land this PR now so that in the meantime we can ensure we dont regress on any of the work that it took to make the shutdown process good enough to make the test pass.

[rukai]

Superceded by #976 and #975
They will provide isolation to the majority of tests, so point in landing this anymore.