fix: guard against null entries in LLM extraction response (#338)

When the LLM returns malformed JSON with null entries in the memories
array (e.g. {memories: [null, {...}]}), accessing raw.category throws:
  TypeError: Cannot read properties of null (reading 'category')

This crash was observed in production (7 occurrences across multiple
days), causing the entire auto-capture pipeline to fail silently.

Add a null/type guard at the top of the candidate processing loop to
skip invalid entries gracefully instead of crashing.

Fixes capture failures logged as:
  memory-lancedb-pro: capture failed: TypeError: Cannot read
  properties of null (reading 'category')

Co-authored-by: dingguagua996-stack <dingguagua996@users.noreply.github.com>

Author: dingguagua996-stack

src/smart-extractor.ts

@@ -401,6 +401,13 @@ export class SmartExtractor {
     let shortAbstractCount = 0;
     let noiseAbstractCount = 0;
     for (const raw of result.memories) {
+      if (!raw || typeof raw !== "object") {
+        invalidCategoryCount++;
+        this.debugLog(
+          `memory-lancedb-pro: smart-extractor: dropping null/invalid candidate entry`,
+        );
+        continue;
+      }
       const category = normalizeCategory(raw.category ?? "");
       if (!category) {
         invalidCategoryCount++;