fix(0089): strip API Gateway stage prefix so /v1 routes resolve (+ complete 0089)

[karczuRF]

The bug (found deploying 0089)

After deploying Prices-production-ApiGateway, every /v1 route returned 404 with a valid API key — /v1/backfill/status, /v1/assets, all of them — while /health returned 200 and a keyless call returned 403.

Root cause: the api-handler Lambda was missing AWS_LAMBDA_HTTP_IGNORE_STAGE_IN_PATH=true, so lambda_http handed axum the path with the API Gateway stage (/production/v1/backfill/status), matching no route → 404 in ~2 ms (no CH call, confirmed in logs).

Why it went unnoticed: /health is a keyless API Gateway mock (never invokes the Lambda), and 0040's tests are all in-process tower::oneshot — this was the first real gateway invocation. It would have 404'd the entire public API in T2, not just /backfill/status.

The fix (config-only, no Rust change)

Add AWS_LAMBDA_HTTP_IGNORE_STAGE_IN_PATH: 'true' to the api-handler environment in compute-stack.ts, mirroring BE's api-handler (BE documents the same var in crates/api/src/common/edge_lock.rs). The Compute redeploy was env-only — diff = one added variable on ApiHandlerFunction, ledger-processor untouched.

Verified live

GET /v1/backfill/status  → 200
{"realtime_tip_ledger":63371525,
 "sdex":{"status":"paused","current_ledger":50457424,"start_ledger":1,"target_ledger":63371525,
         "progress_pct":79.62,"ledgers_remaining":12914101,"last_push_at":"2026-07-07T19:37:03Z"},
 "soroban_amm":{"status":"running","last_push_at":"2026-07-07T19:37:03Z","completed_at":null}}
no-key → 403,  /health → 200

Also completes + archives task 0089 — the M1 AC "GET /backfill/status endpoint live and returning valid progress data" is now satisfied.

M1 after this

Remaining M1: 0088 (backfill run, paused) + 0082 (worker/MV verification). The read API is now live.