Parallelize file linting with n-par-map

Use Guile's n-par-map from (ice-9 threads) to lint files in parallel,
one thread per CPU core. Results are collected per-file then output
sequentially to maintain deterministic ordering.

Performance on Guix tree (1336 files, 12 cores):

  Before (sequential)     After (parallel)
  full:    21s             full:    14s   (1.5x)
  surface: 15s             surface:  8s   (1.9x)
  line:    14s             line:     7s   (2.0x)

gulie/engine.scm

@@ -7,6 +7,7 @@
 (define-module (gulie engine)
   #:use-module (ice-9 ftw)
   #:use-module (ice-9 textual-ports)
+  #:use-module (ice-9 threads)
   #:use-module (srfi srfi-1)
   #:use-module (gulie diagnostic)
   #:use-module (gulie rule)
@@ -127,14 +128,22 @@ and '%min-severity with value 'error, 'warning, or 'info (default)."
     (sort diagnostics diagnostic<?)))
 
 (define (lint-files files config)
-  "Lint multiple FILES. Returns total diagnostic count."
-  (let ((total 0))
+  "Lint multiple FILES in parallel. Returns total diagnostic count.
+Uses n-par-map to distribute work across threads, then outputs
+diagnostics sequentially to maintain deterministic file order."
+  (let* ((ncpus (max 1 (total-processor-count)))
+         (results (n-par-map ncpus
+                             (lambda (file)
+                               (cons file (lint-file file config)))
+                             files))
+         (total 0))
+    ;; Output in original file order (n-par-map preserves it)
     (for-each
-     (lambda (file)
-       (let ((diags (lint-file file config)))
+     (lambda (result)
+       (let ((diags (cdr result)))
          (set! total (+ total (length diags)))
          (format-diagnostics diags (current-output-port))))
-     files)
+     results)
     total))
 
 (define (scheme-file? path)