Add OpenTelemetry metrics for job callbacks

[epompeii]

Callbacks get OpenTelemetry metrics.

• callback.attempt counts each attempt by its class: a status class, a timeout, a connection error, or a block with its reason.
• callback.finish.duration and callback.finish.attempts record each callback that settles, by its outcome and failure reason: its latency from the start of its delivery, and the attempts it made over its life. Only a settle that committed is recorded, and both have bucket boundaries sized for their values.
• callback.pending is an observable gauge of the callbacks waiting to be delivered, read from the database at each export on a read connection only if one is free, so a busy pool skips a point instead of stalling the export.

Every attribute comes from a fixed enum, so no attribute carries a URL, a header, an address, or any other value from the request.

Needed for #1046

[github-actions]

Bencher Report

Project	Bencher
Branch	u/ep/callback/metrics
Testbed	intel-v1

Click to view all benchmark results

Benchmark	Latency	Benchmark Result microseconds (µs) (Result Δ%)	Upper Boundary microseconds (µs) (Limit %)
Adapter::Json	📈 view plot 🚷 view threshold	5.17 µs (+4.95%) Baseline: 4.92 µs	6.00 µs (86.20%)
Adapter::Magic (JSON)	📈 view plot 🚷 view threshold	4.97 µs (+4.34%) Baseline: 4.76 µs	5.68 µs (87.42%)
Adapter::Magic (Rust)	📈 view plot 🚷 view threshold	27.63 µs (+3.58%) Baseline: 26.67 µs	30.32 µs (91.11%)
Adapter::Rust	📈 view plot 🚷 view threshold	4.68 µs (+15.48%) Baseline: 4.06 µs	6.35 µs (73.70%)
Adapter::RustBench	📈 view plot 🚷 view threshold	4.66 µs (+15.06%) Baseline: 4.05 µs	6.33 µs (73.61%)

🐰 View full continuous benchmarking report in Bencher