fix: address bot review — surrogate-pair counting and CJK line splitting

- Use code-point length instead of UTF-16 length in estimateStringChars()
  so that CJK Extension B+ surrogate pairs (U+20000+) are counted as 1
  character, not 2 (fixes ~25% overestimate for rare characters).

- Change long-line split step from maxChars to chunking.tokens so that
  CJK lines are sliced into token-budget-sized segments instead of
  char-budget-sized segments that produce ~4x oversized chunks.

- Add tests for both fixes: surrogate-pair handling and long CJK line
  splitting.

Addresses review feedback from Greptile and Codex bots.

packages/memory-host-sdk/src/host/internal.ts

@@ -398,8 +398,13 @@ export function chunkMarkdown(
     if (line.length === 0) {
       segments.push("");
     } else {
-      for (let start = 0; start < line.length; start += maxChars) {
-        segments.push(line.slice(start, start + maxChars));
+      // Use token count (not maxChars) as the split step so that CJK lines
+      // – where 1 char ≈ 1 token – are sliced into budget-sized segments.
+      // For Latin text the token count is ≥ maxChars/4, which still produces
+      // segments well within the char budget after weighting.
+      const splitStep = Math.max(1, chunking.tokens);
+      for (let start = 0; start < line.length; start += splitStep) {
+        segments.push(line.slice(start, start + splitStep));
       }
     }
     for (const segment of segments) {

src/utils/cjk-chars.test.ts

@@ -81,7 +81,6 @@ describe("estimateStringChars", () => {
     // "你" counts as 4, emoji remains 2 => total 6
     expect(estimateStringChars("你😀")).toBe(6);
   });
-
   it("yields ~1 token per CJK char when divided by CHARS_PER_TOKEN_ESTIMATE", () => {
     // 10 CJK chars should estimate as ~10 tokens
     const cjk = "这是一个测试用的句子呢";