feat(replacer): expose --max-poll-interval-ms and --health-check-file into the CLI

[aldy505]

Currently there are no way in self-hosted to be able to configure many things on snuba replacer. This PR adds --max-poll-interval-ms and --health-check-file.

[sentry]

Bug: The default value for --max-poll-interval-ms unconditionally overrides the Kafka consumer's max.poll.interval.ms, reducing it from 5 minutes to 30 seconds for all deployments.
_{Severity: HIGH}

Suggested Fix

Change the default value of the max_poll_interval_ms argument to None. Only apply the max.poll.interval.ms setting to the Kafka consumer configuration if the value is not None, following the existing pattern in snuba/consumers/consumer_builder.py.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's
not valid.

Location: snuba/cli/replacer.py#L113-L117

Potential issue: The `replacer` CLI unconditionally applies a default
`max.poll.interval.ms` of 30000ms (30 seconds) to the Kafka consumer configuration. This
overrides `librdkafka`'s default of 300000ms (5 minutes) for all existing deployments
that do not specify the new `--max-poll-interval-ms` flag. Because `session.timeout.ms`
must be less than or equal to this value, it is also reduced. If a replacement operation
takes longer than 30 seconds, which is common for large replacements, the consumer will
be removed from its consumer group, leading to rebalancing loops in production.

_{Did we get this right? 👍 / 👎 to inform future reviews.}

[sentry]

Bug: Setting session.timeout.ms equal to max_poll_interval_ms is risky, as it leaves no buffer for processing delays and can cause frequent, unnecessary consumer rebalances.
_{Severity: MEDIUM}

Suggested Fix

Decouple session.timeout.ms from max_poll_interval_ms. Set session.timeout.ms to a static, reasonable value (e.g., the default 45000ms) that is independent of max_poll_interval_ms, while still ensuring the constraint session.timeout.ms <= max_poll_interval_ms is met. This provides a buffer for processing and prevents instability from transient delays.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's
not valid.

Location: snuba/cli/replacer.py#L116-L117

Potential issue: The code sets the Kafka consumer's `session.timeout.ms` to be equal to
`max_poll_interval_ms` when the latter is below 45 seconds. This conflates two distinct
timeouts: session liveness and message processing time. By making them equal (e.g., at
the default of 30 seconds), there is no tolerance for transient delays like GC pauses,
network jitter, or I/O latency. This can cause the Kafka broker to prematurely consider
the consumer dead, triggering frequent and unnecessary group rebalances, which degrades
performance and stability. This pattern is explicitly marked as a "HACK" in another part
of the codebase.

Also affects:

• snuba/consumers/consumer_builder.py:175-176