provider: reprocess(recordingId, job, model) against /reprocess

Server detail message surfaces through ProviderError.Transient so 409
reasons ('already running', 'transcribe first') reach the UI.

android/app/src/main/java/com/shonar/provider/CustomShonarProvider.kt

@@ -306,6 +306,35 @@ class CustomShonarProvider(
         "/api/v1/recordings/$recordingId/jobs", recordingId
     )
 
+    /** Force one pipeline stage to re-run on the server (no re-upload).
+     *  [model] only applies to job="transcribe" (switches the saved
+     *  per-recording override). 409 = stage already running or nothing to
+     *  summarize; surfaced as ProviderError.Transient with the server's
+     *  detail message. */
+    suspend fun reprocess(recordingId: String, job: String, model: String? = null): String =
+        withContext(Dispatchers.IO) {
+            ensureConnected()
+            val base = currentOrigin()
+            val q = if (model != null) "&model=$model" else ""
+            executeAuthed(base) { token ->
+                post(base, "/api/v1/recordings/$recordingId/reprocess?job=$job$q", token, "{}")
+            }.use { resp ->
+                when (resp.code) {
+                    200, 201 -> resp.body?.string().orEmpty()
+                    404 -> throw ProviderError.NotFound(recordingId)
+                    else -> {
+                        val body = runCatching {
+                            org.json.JSONObject(resp.body?.string().orEmpty())
+                                .optString("detail")
+                        }.getOrNull()
+                        throw ProviderError.Transient(
+                            body?.takeIf { it.isNotBlank() }
+                                ?: "Reprocess failed (HTTP ${resp.code})")
+                    }
+                }
+            }
+        }
+
     /** User edit → new server version with edited_by_user=true (pipeline won't clobber). */
     suspend fun updateTranscript(
         recordingId: String,