feat(resume): admin bulk resume import

Adds the dashboard surface for the backend's /api/admin/v1/resume/import-batches
endpoints, and the multipart support it needed.

Client:
- rawRequest passes a FormData body through untouched and skips Content-Type,
  so the browser sets multipart/form-data *with the boundary*. Setting it by
  hand stripped the boundary and the server saw an empty upload — verified
  against the backend: 202 with the fix, 422 VALIDATION_ERROR without it.
  This was the first admin endpoint in the codebase to accept a file.
- New api.upload(). fetch reports no upload progress, so upload UI stays
  indeterminate rather than faking a percentage.

UI (/resume/import-batches, gated on admin:resume:read; writes on :manage):
- Upload dialog with countryCode and dry-run checked by default — seeing the
  collision report before creating accounts is almost always what you want.
- Detail sheet polls every 2s and stops once the batch is terminal. Clickable
  count tiles filter the per-resume report; step statuses surface `命中缓存` /
  `复用结果` so the dedup is visible.
- Confirm a dry run, retry failed rows, and resolve a missing contact.
  IMPORT_IDENTITY_TAKEN is shown inline — a collision is expected input, not
  an error toast.
- progress.total is 0 until unpacking finishes, so both views render 准备中…
  rather than "0 / 0", which would read as an empty batch.

types.gen.ts regenerated from the backend's admin spec.

Co-Authored-By: Claude Opus 5 <[email protected]>

src/api/types.ts

@@ -738,3 +738,89 @@ export interface ExcProvider {
     created_at?: string
   }>
 }
+
+// ---------- resume bulk import ----------
+
+export type ResumeImportBatchStatus =
+  | 'queued'
+  | 'unpacking'
+  | 'processing'
+  | 'completed'
+  | 'failed'
+
+/**
+ * What happened to one resume. Independent of the parse status: a resume can
+ * parse perfectly and still end up `skipped_duplicate`.
+ */
+export type ResumeImportOutcome =
+  | 'created'
+  | 'would_create'
+  | 'skipped_duplicate'
+  | 'reused_import'
+  | 'needs_contact'
+  | 'duplicate_file'
+  | 'unsupported_entry'
+  | 'entry_too_large'
+  | 'parse_failed'
+  | 'failed'
+
+/** `cached` / `reused` mean no provider call was made for that step. */
+export type ResumeImportStepStatus =
+  | 'pending'
+  | 'processing'
+  | 'completed'
+  | 'cached'
+  | 'reused'
+  | 'skipped'
+  | 'failed'
+
+export interface ResumeImportBatch {
+  id: string
+  kind: 'resume-import-batch'
+  status: ResumeImportBatchStatus
+  dryRun: boolean
+  progress: {
+    /** 0 until unpacking finishes — render a preparing state, not "0 / 0". */
+    total: number
+    processed: number
+    percent: number
+  }
+  counts: Partial<Record<ResumeImportOutcome, number>>
+  archive: { filename: string | null; size: number | null }
+  options: {
+    countryCode: string | null
+    languageHint: string | null
+    schemaVersion: string | null
+  }
+  error: { code: string; message: string } | null
+  createdBy: string | null
+  pollUrl: string
+  createdAt: string | null
+  startedAt: string | null
+  finishedAt: string | null
+}
+
+export interface ResumeImportItem {
+  id: string
+  kind: 'resume-import'
+  status: 'queued' | 'processing' | 'completed' | 'failed'
+  outcome: ResumeImportOutcome | null
+  steps: { extraction: ResumeImportStepStatus; structure: ResumeImportStepStatus }
+  document: {
+    filename: string | null
+    archiveEntry: string | null
+    mimeType: string | null
+    size: number | null
+    extractionMethod: string | null
+  }
+  identityType: 'phone' | 'email' | null
+  matchedUserUid: string | null
+  duplicateOfImportId: string | null
+  error: { code: string; message: string } | null
+  createdAt: string | null
+  finishedAt: string | null
+  // detail-only — never present on the list
+  identityValue?: string | null
+  metrics?: Record<string, number> | null
+  result?: Record<string, unknown> | null
+}