feat(solana-driver): Solver interaction

[tilacog]

Description

Adds solver interaction for the solana-driver crate.

Changes

• New Solver type that sends POST /solve requests to configured solvers. It is now part of the API State struct.
• Added a minimal set of domain types to support this interaction: Auction, Order, OrderUID, Solution, etc.
  • Each with a mininmal set of fields too, because right now they serve just to produce the DTO set for the solver.
  • I'll expand on them as we integrate w/ the autopilot and with the Settlement program.
• Added a set of DTO definitions to support this interaction, mirroring the solana-solvers DTO for the Jupiter API and Solana SDK types returned by it.
• Small refactor on the solana-solvers axum API to split bind and serve, so we can import the solver process and run it in integration tests (following the workspace convention).
• Added a new URL deserialization helper fn in serde-ext to prevent URLs without a trailing slash to be set in config (after this bot review comment)

How to test

New live integration test: driver_solves_against_live_jupiter_engine, run it with:

$ cargo nextest run -p solana-driver --run-ignored ignored-only --test jupiter_live

[claude]

Claude finished @tilacog's task in 2m 19s —— View job

---

PR Review: Solver interaction for solana-driver

Reviewed the solver client, domain/DTO types, API/config wiring, and tests. Overall this is clean, well-documented, and well-tested — the domain/DTO split with serde tests pinning the wire shape against the solver crate is a nice touch, and the live Jupiter test does a thorough end-to-end check of the buy_destination derivation.

I left 3 inline comments:

1. solver/mod.rs — timeout computed before acquiring the semaphore permit. Queue-wait time isn't deducted from the deadline budget, so a queued solve can run past the auction deadline. Suggest acquiring the permit first, then computing remaining.
2. dto/solution.rs — into_domain fails the whole response on one bad trade. A single unknown order UID short-circuits both collect()s and discards every solution from that engine, which contradicts the doc comment ("makes its solution invalid"). Either filter per-solution or update the doc.
3. solver/mod.rs — Url::join("solve") footgun for endpoints with a path prefix (low priority; current config is fine).

Nothing blocking — (1) and (2) are worth a look before merge.

[squadgazzz]

@tilacog , please resolve conflicts. Otherwise, the diff is confusing.

[tilacog]

@squadgazzz I've reconciled this branch w/ latest main, should be good for a review now.

[squadgazzz]

LGTM