Clarify ignore_unhealthy_builders behavior for unhealthy builders

[nogie-dev]

Finding: Clarify ignore_unhealthy_builders semantics

Affected Path

• crates/rollup-boost/src/cli.rs
• crates/rollup-boost/src/server.rs
• crates/rollup-boost/src/tests/unhealthy_builder_traffic.rs

Trigger

The ignore_unhealthy_builders flag appears to have ambiguous semantics across its CLI help text, implementation, and test comments.

The CLI help says:

/// Allow all engine API calls to builder even when marked as unhealthy
#[arg(long, env, default_value = "false")]
pub ignore_unhealthy_builders: bool,

This reads as: when the flag is true, rollup-boost should ignore the unhealthy health state and continue sending builder Engine API calls.

The current implementation skips builder traffic when the flag is true and the builder is unhealthy:

fn should_skip_unhealthy_builder(&self) -> bool {
    self.ignore_unhealthy_builders && !matches!(self.probes.health(), Health::Healthy)
}

The current unhealthy-builder traffic test also configures
ignore_unhealthy_builders=true, but later comments/assertions describe the scenario as flag=false.

Impact

The observable behavior may be surprising for operators:

• ignore_unhealthy_builders=false currently means unhealthy builders still receive Engine API calls.
• ignore_unhealthy_builders=true currently means unhealthy builders areskipped.

That seems opposite to the CLI help text if "ignore" refers to ignoring the unhealthy health state.

Suggested fix

I think the cleanest fix would be to first clarify whether ignore_unhealthy_builders is meant to ignore the unhealthy status and continue sending builder calls, or to ignore/skip unhealthy builders. Once that intended behavior is confirmed, the predicate, CLI help text, and test comments can be aligned in a small follow-up PR.