Add a test which verifies that a single "broken" ingester is not affecting read path

[simonswine]

I have noticed a case where a single timing out ingester can affect the availability of the read path.

I would like to verify this behaviour by using an integration test.

In the particular example I have noticed the MergeProfilesStacktraces timed out, after a kubernetes node failure. The ingester status was correctly recongnized as unhealthy.

[simonswine]

An internal slack reference for more details for GL employees: https://raintank-corp.slack.com/archives/C04TRP742NN/p1709718028724059