fix(services): retry malformed capability probes after the existing backoff

Rebased onto current main with review follow-ups folded in.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

src/services/requests.js

@@ -66,9 +66,9 @@ export function createApplicationRequestServices({
       throw new Error(`${label} unavailable (${response.status})`);
     return response.data;
   }
-  // Learn once per page whether the server has an Overpass instance, so an
+  // Remember valid capability decisions for the page, so an
   // unconfigured server is never asked (and never answers with an error).
-  // Discovery is bounded; a failed or timed-out probe is forgotten and not
+  // Discovery is bounded; failed, malformed or timed-out probes are forgotten and not
   // retried for a short backoff, during which queries go to the server as
   // they would against an older server without the probe.
   let boundaryCapability = null;
@@ -79,10 +79,13 @@ export function createApplicationRequestServices({
     boundaryCapability = request(`${urls.boundaries}/status`, {
       signal: AbortSignal.timeout(boundaryProbe.timeoutMs),
     }).then(
-      (response) =>
-        response.ok && typeof response.data?.configured === 'boolean'
-          ? response.data.configured
-          : null, // an older server: fall back to asking per query
+      (response) => {
+        if (response.ok && typeof response.data?.configured === 'boolean')
+          return response.data.configured;
+        boundaryCapability = null;
+        boundaryProbeRetryAt = Date.now() + boundaryProbe.retryMs;
+        return null; // capability unknown: let this query ask the server
+      },
       (error) => {
         boundaryCapability = null;
         if (lifetime?.aborted) throw error;