fix(kura): derive FD pool from process capacity

Resolves https://github.com/tuist/tuist/issues/11132

This adjusts Kura’s file descriptor admission control for bursty REAPI ByteStream uploads:

• Raise the soft RLIMIT_NOFILE before loading config so dynamic defaults are derived from the real process capacity.
• Derive the internal FD pool and segment handle cache from the detected limit while keeping bounded headroom and the existing 5s acquire timeout.
• Stop pinning Helm’s default FD pool and segment cache values so production deployments use dynamic sizing unless explicitly overridden.
• Return RESOURCE_EXHAUSTED for REAPI ByteStream writes when the FD pool is exhausted, instead of surfacing it as an internal CAS persistence failure.

How to test locally

• mise exec -- cargo fmt
• git diff --check
• mise exec -- cargo test config::tests --lib
• mise exec -- cargo test reapi::tests::bytestream_writes_persist_completely_under_concurrency --lib
• helm template kura ./kura/ops/helm/kura
• helm template kura ./kura/ops/helm/kura --set config.fileDescriptors.poolSize=2048 --set config.fileDescriptors.segmentHandleCacheSize=128

No GitHub comments yet.