solana-driver: Autopilot interface

[tilacog]

Description

Introduces the HTTP interface for receiving /solve and /settle requests from the autopilot.

The /settle endpoint is a stub for now because the actual settlement logic will be presented in a follow-up PR, but all the other aspects of it's API are in place.

@squadgazzz: This PR diff ended up a bit bigger than I expected. Please let me know if you prefer me to split it into smaller ones. (I'm taking care this doesn't repeat for the follow-up PRs)

Changes

• crates/autopilot-svm: SolveRequest, Solution, SettleRequest now serialize integer ids as decimal strings, given u64 can't fit in a JSON number.
• Updated domain types:
  • Competition (new type, holds a list of Solvers and orchestrates solve calls among them.
  • Auction: its id field, rejects non-positive values
  • Order restructured fields to match what we get from the autopilot.
• implemented api routing for solve and settle.
  • Also defined api-specific Error and error::Kind for invalid data (auction ids, solver failures, etc)

How to test

cargo test -p solana-driver

New TODOs

These are the new TODO comments introduced by this PR.
We're shipping a basic version of the solana-driver first and will optimize these areas later.

• In competition.rs, the work fan-out is very simple/naive and there's no solution cache.
• The autopilot sends a deadline_slot that the driver needs to resolve into an actual timestamp, but for now this PR just uses a hardcoded 15s deadline for every auction. (The deadline_slot field is included in the DTO, just not enforced.)
• The solana-solver engine doesn't yet report the counterpart (buy/sell) amount, so the driver stubs the score and counterpart amount to 0.

[claude]

Claude finished @tilacog's task in 2m 43s —— View job

---

PR Review: solana-driver Autopilot interface

Solid, well-documented PR — the mirrored DTOs with pinned-literal tests against autopilot-svm are a nice way to keep the two crates in sync, and the acknowledged TODOs (naive fan-out, hardcoded 15s deadline, stubbed score/counterpart amount) are clearly scoped. I confirmed the .expect() in solve_response.rs is safe by construction since Solution::into_domain rejects unknown order UIDs. A few points worth a change, posted inline:

• SolverFailed → HTTP 400 (error.rs): all-engines-failed is an internal driver failure, not a bad request; should be 500.
• /settle now panics (settle/mod.rs): a reachable HTTP path aborting the connection is worse than the previous clean 501.
• Id(pub i64) (auction.rs): public field bypasses the positivity invariant (minor).

Nothing blocking beyond those; the first two are quick fixes.

• Read PR diff and changed files
• Analyze domain, api, dto changes
• Post inline comments for issues
• Post summary

[squadgazzz]

LGTM