Handle Dynamo request-priority headers in Pylon

[along-2017]

Description

Dynamo schedules on x-dynamo-request-priority (i32, higher value is more important) and silently ignores values that do not parse as i32. The platform resolves a caller priority and carries it as x-priority (u32, lower value is more urgent, absent means unconfigured).

Make Pylon the single component that handles the engine-facing Dynamo headers: it derives x-dynamo-request-priority from x-priority when forwarding to the upstream inference server, and sanitizes inbound values. The gateway and Stargate stay Dynamo-agnostic, so if the Dynamo header contract changes later (names, polarity, types), only Pylon changes.

Definition of Done

• Pylon removes any inbound x-dynamo-request-* header before forwarding upstream, so client-supplied values never reach the engine.
• Pylon emits x-dynamo-request-priority only when the inbound x-priority header is present. Absent stays absent: the parsed default of 0 is never inverted into a maximum engine priority.
• The mapping inverts polarity and always produces an i32-valid value (for example i32::MAX - min(x, i32::MAX)), so no configured priority is silently dropped by Dynamo.
• Pylon logs the emitted value with the request context for per-request triage.
• No Dynamo-specific header handling is added to the gateway or Stargate.
• Unit tests cover the mapping boundaries (0, i32::MAX, u32::MAX), the presence rule, and the inbound strip; an integration test pins end-to-end delivery using the mock-dynamo header recording.

Resources

• Pylon forwarding path: src/libraries/rust/stargate/crates/pylon-lib/src/quic_http_tunnel/core.rs
• Header parsing: src/libraries/rust/stargate/crates/pylon-lib/src/request_observer/headers.rs
• Dynamo priority headers: https://github.com/ai-dynamo/dynamo/blob/main/docs/components/frontend/nvext.md