chore(compute): Minor compute_ctl startup refactoring

[ololobus]

Problem

While debugging neondatabase/cloud#24882 I noticed that there is a significant amount of startup time (hundreds of ms) that is not attributed to any particular phase.

The current code structure makes it harder to add additional per-startup phase metrics.

Summary of changes

Refactor compute startup:

1. Between getting spec from control plane and calling start_compute() we do not have any meaningful code except launching some aux threads.
2. Move swap and disk quota configuration to a separate configure_vm() compute helper.
3. Do not try to launch vm-monitor if Postgres startup failed and we are exiting anyway.

PR with more metrics will follow up, I do not want to pack too many changes in a single one.

[hlinnaka]

@ololobus only noticed this now. I was working on similar refactorings, see PR #11007. I went much further, but it's the same general direction I think. I propose that we proceed with #11007. But if some issue with that arises in review or CI, we could fall back to this one.

[hlinnaka]

PR with more metrics will follow up, I do not want to pack too many changes in a single one.

I'm curious to see what you have in mind. I've been playing with the OpenTelemetry traces again. It's great because it gives you a very clear visualization of where exactly the time is spent, but requires a bit of setup to collect the traces.

[github-actions]

7744 tests run: 7364 passed, 0 failed, 380 skipped (full report)

---

Flaky tests (3)

Postgres 17

• test_migration_to_cold_secondary: release-x86-64-with-lfc, release-x86-64-without-lfc, release-arm64-without-lfc

Code coverage* (full report)

• functions: 32.8% (8636 of 26362 functions)
• lines: 48.6% (73085 of 150452 lines)

* collected from Rust tests only

---

_{The comment gets automatically updated with the latest test results
6cfaa77 at 2025-02-26T21:57:39.647Z :recycle:}