feat(compute_ctl): run lease lsn in tokio and in parallel

[skyzh]

Problem

Part of LKB-2634

The main motivation of this patch is to make the lease LSN requests in parallel. On that, it seems like a refactor to make it running in async is a good choice.

Summary of changes

• Issue lease_lsn requests in parallel.
• ..which is done by futures::stream::buffered_unordered.
• revert the test case test_readonly_node_gc to issue lease requests only from the compute

[github-actions]

9075 tests run: 8402 passed, 0 failed, 673 skipped (full report)

---

Flaky tests (1)

Postgres 17

• test_partial_evict_tenant[relative_equal]: release-arm64-with-lfc

Code coverage* (full report)

• functions: 34.7% (8840 of 25457 functions)
• lines: 45.8% (71633 of 156461 lines)

* collected from Rust tests only

---

_{The comment gets automatically updated with the latest test results
5622bf0 at 2025-07-30T20:02:06.957Z :recycle:}

[skyzh]

need to asyncify wait_timeout_while_pageserver_connstr_unchanged, converting to draft

[skyzh]

cc @alexanderlaw where can I trigger 100 runs of a specific test? I'd like to see if the previous flakyness can be fixed without changing the test case. Thanks!

[alexanderlaw]

cc @alexanderlaw where can I trigger 100 runs of a specific test? I'd like to see if the previous flakyness can be fixed without changing the test case. Thanks!

Please try something like this:

gh workflow run 'Build and Run Selected Test' --ref skyzh/stable-test-readonly-node-gc -f test-selection='test_readonly_node_gc' -f run-count='30' -f  archs='["x64", "arm64"]' -f build-types='["release"]' -f pg-versions='[{"pg_version":"v14"}, {"pg_version":"v15"}, {"pg_version":"v17"}]'

[erikgrinaker]

This needs spawn_blocking for wait_timeout_while_pageserver_connstr_unchanged too.

[erikgrinaker]

I think it's fine to just use an unbounded FuturedUnordered here. If the tenant has many shards, they are mostly likely running on many Pageservers, so the concurrency limiting doesn't really do anything for the server. For the client side, we don't care about opening O(16) concurrent connections.