op-conductor,op-node: allow system to select port, make op-node wait for conductor endpoint

[protolambda]

Description

• Make conductor actually listen on the host that it is configured to, rather than always 0.0.0.0
• Make conductor optionally default to the address it binds to, rather than a preconfigured address, for advertising their self as peer
• Add conductor CLI option to change the server-address it advertises itself as, to not only use the same address it binds to, in case you bind it to an IP that doesn't resolve otherwise. If unspecified, it defaults to what address it is listening on.
• Make op-node wait for the conductor RPC to be available. This is instantly available in regular operation, from CLI. But may take longer in case of an e2e test, where the conductor is started after the op-node. The conductor RPC client is lazy-loaded (initialize already runs on usage of the RPC method, rather than creation of the RPC client).

Tests

• E2E system now sets up conductors without trying to reserve ports. This prevents flakes where endpoints aren't actually reserved, or where endpoints are not up and running yet.

Note: TestSequencerFailover_ActiveSequencerDown still seems to flake due to a leadership transfer timeout. Maybe because of the 1-second timeout, which is quite low for CI on limited resources. So I bumped the conductor RPC timeout to 5 seconds.

And I am leaving the op-conductor logging on debug level, so we can look at any further flakes if they do happen.

[mslipper]

looks good to me, but would like someone with more conductor experience to opine before merging.

[tynes]

I think this code should be moved over to the infra repo given the infra team owns the code in the infra repo and this repo is mostly owned by the protocol team

[axelKingsley]

[axelKingsley]

(My bad, I misclicked when reading this review and closed it momentarily)

[zhwrd]

one change suggestion, otherwise lgtm

[zhwrd]

I think this code should be moved over to the infra repo given the infra team owns the code in the infra repo and this repo is mostly owned by the protocol team

Not sure its about what team owns what code, imo we really need e2e tests for conductor to run when op-stack changes. We actually need to expand the e2e test suite for conductor since we've had a few regressions in op-stack introduced that could be caught earlier. I'm actually also considering versioning conductor with the rest of the op-stack since they are pretty tightly coupled, but tbd