fix: Grammar retry on rate-limit + cursor position after correction

- Grammar service: transient errors (rate limiting, timeout) no longer cached permanently — next request will retry loading instead of failing forever
- Editor: cursor now placed right after corrected text instead of jumping to start — applied to all 3 code paths (UUID span, text-match fallback, offset fallback)

src/js/editor.js

@@ -451,6 +451,15 @@ function applySuggestionAtOffsets(suggestion) {
       parent.insertBefore(correctedNode, errorSpan);
       parent.removeChild(errorSpan);
       parent.normalize();
+      // Place cursor right after the corrected text
+      try {
+        const sel = window.getSelection();
+        const r = document.createRange();
+        r.setStartAfter(correctedNode);
+        r.collapse(true);
+        sel.removeAllRanges();
+        sel.addRange(r);
+      } catch(e) {}
     } else {
       // Fallback: find span by matching original text
       const allErrorSpans = document.querySelectorAll('.spelling-error, .grammar-error, .punctuation-suggestion');
@@ -462,6 +471,15 @@ function applySuggestionAtOffsets(suggestion) {
           p.insertBefore(correctedNode, span);
           p.removeChild(span);
           p.normalize();
+          // Place cursor right after the corrected text
+          try {
+            const sel = window.getSelection();
+            const r = document.createRange();
+            r.setStartAfter(correctedNode);
+            r.collapse(true);
+            sel.removeAllRanges();
+            sel.addRange(r);
+          } catch(e) {}
           found = true;
         }
       });
@@ -472,6 +490,8 @@ function applySuggestionAtOffsets(suggestion) {
         const after = text.substring(suggestion.end);
         const newText = before + suggestion.correction + after;
         setEditorHTML(escapeHtml(newText));
+        // Place cursor after the inserted correction
+        setCaretOffset(suggestion.start + suggestion.correction.length);
       }
     }
     hideTooltip();
@@ -517,6 +537,15 @@ function applyAlternativeCorrection(suggestion, correctionText) {
       parent.insertBefore(correctedNode, errorSpan);
       parent.removeChild(errorSpan);
       parent.normalize();
+      // Place cursor right after the corrected text
+      try {
+        const sel = window.getSelection();
+        const r = document.createRange();
+        r.setStartAfter(correctedNode);
+        r.collapse(true);
+        sel.removeAllRanges();
+        sel.addRange(r);
+      } catch(e) {}
     } else {
       const allErrorSpans = document.querySelectorAll('.spelling-error, .grammar-error, .punctuation-suggestion');
       let found = false;
@@ -527,6 +556,15 @@ function applyAlternativeCorrection(suggestion, correctionText) {
           p.insertBefore(correctedNode, span);
           p.removeChild(span);
           p.normalize();
+          // Place cursor right after the corrected text
+          try {
+            const sel = window.getSelection();
+            const r = document.createRange();
+            r.setStartAfter(correctedNode);
+            r.collapse(true);
+            sel.removeAllRanges();
+            sel.addRange(r);
+          } catch(e) {}
           found = true;
         }
       });
@@ -536,6 +574,8 @@ function applyAlternativeCorrection(suggestion, correctionText) {
         const after = text.substring(suggestion.end);
         const newText = before + correctionText + after;
         setEditorHTML(escapeHtml(newText));
+        // Place cursor after the inserted correction
+        setCaretOffset(suggestion.start + correctionText.length);
       }
     }
     hideTooltip();

src/nlp/grammar/grammar_service.py

@@ -67,6 +67,9 @@ def get_grammar_model():
     """
     Lazy-load the grammar model on first call.
     Returns the GrammarChecker instance, or raises RuntimeError if loading fails.
+    
+    Transient errors (rate limiting, network timeouts) are NOT cached —
+    the next request will retry loading. Only permanent failures are cached.
     """
     global _grammar_checker, _load_error
 
@@ -101,9 +104,22 @@ def get_grammar_model():
 
     except Exception as e:
         import traceback
-        _load_error = str(e)
+        error_msg = str(e)
         logger.error(f"Failed to load grammar model: {e}")
         logger.error(traceback.format_exc())
+
+        # Transient errors (rate limiting, network) should NOT be cached —
+        # allow retry on next request
+        transient_keywords = ['Too many requests', 'rate limit', 'timeout',
+                              'ConnectionError', 'ConnectTimeout', 'ReadTimeout']
+        is_transient = any(kw.lower() in error_msg.lower() for kw in transient_keywords)
+
+        if is_transient:
+            logger.warning(f"Grammar load error is TRANSIENT — will retry on next request: {error_msg}")
+            # Do NOT set _load_error — next call will retry
+        else:
+            _load_error = error_msg  # Cache permanent failures only
+
         raise RuntimeError(f"Grammar model load failed: {e}")