Add interface for collecting performance metrics

[amomchilov]

E.g. to emit to StatsD / OpenTelemetry how long each pipeline stage took.

[vinistock]

Worth checking out https://crates.io/crates/tracing

[st0012]

I had a quick investigation with River about OTel integration with Observe and here's a quick summary why it's not worth the effort for now:

• Integration. Ruby LSP and rubydex (in-process Rust extension) each need
  their own OTel SDK, exporter, and flush-on-exit — they can't share one. That's
  two instrumentation stacks plus per-call traceparent propagation across the
  FFI boundary to keep both halves in one trace.
• Emitting to Observe from local machines. Production exports to an in-cluster
  collector; a laptop can't reach it. Traces must flow through the observe-dev
  local collector (which most devs won't have running), the ruby-lsp service
  name needs allowlisting in the ingestion pipeline, and the Observability team
  has to confirm laptop-originated LSP traces are supported. Until that's settled,
  the data may not even land.
• Extra dependencies. Adds the opentelemetry-shopify gem on the Ruby side and
  the upstream opentelemetry / opentelemetry_sdk / opentelemetry-otlp /
  tracing-opentelemetry crates on the Rust side, all to maintain and keep
  version-aligned.

It'd be amazing to be able to see a waterfall breakdown for Rubydex indexing traces like below, but it's just too much work.