fix: UTF-8 char boundary panics in smart chunker + schema migration order

Two bugs found during real vault testing:

1. Smart chunker panicked on multi-byte UTF-8 chars (em dash, etc.)
   when byte offsets from break-point scoring landed inside multi-byte
   sequences. Fixed by snapping all byte offsets to valid char
   boundaries before slicing.

2. Schema migration failed on existing v0.1 databases: the SCHEMA
   constant tried to CREATE INDEX on docid column before migration
   added it. Moved index creation into the migration path so it
   runs after the column exists.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: devwhodevs

src/chunker.rs

@@ -163,6 +163,16 @@ fn approx_tokens(text: &str) -> usize {
     text.len().div_ceil(4)
 }
 
+/// Snap a byte offset to the nearest valid UTF-8 char boundary (forward).
+fn snap_to_char_boundary(s: &str, offset: usize) -> usize {
+    let offset = offset.min(s.len());
+    let mut pos = offset;
+    while pos < s.len() && !s.is_char_boundary(pos) {
+        pos += 1;
+    }
+    pos
+}
+
 /// Extract the first heading line from text (any `#` level).
 fn extract_heading(text: &str) -> Option<String> {
     for line in text.lines() {
@@ -205,6 +215,10 @@ pub fn smart_chunk(content: &str, target_tokens: usize, overlap_pct: usize) -> V
     let mut start_offset = 0;
 
     while start_offset < content.len() {
+        start_offset = snap_to_char_boundary(content, start_offset);
+        if start_offset >= content.len() {
+            break;
+        }
         let remaining = &content[start_offset..];
         if remaining.trim().is_empty() {
             break;
@@ -252,7 +266,10 @@ pub fn smart_chunk(content: &str, target_tokens: usize, overlap_pct: usize) -> V
             Some(bp) => bp.byte_offset,
             None => {
                 // No good break point found; cut at target
-                let cut = (start_offset + target_chars).min(content.len());
+                let cut = snap_to_char_boundary(
+                    content,
+                    (start_offset + target_chars).min(content.len()),
+                );
                 // Try to find a newline near the cut
                 let fallback = if let Some(nl) = content[start_offset..cut.min(content.len())]
                     .rfind('\n')
@@ -267,6 +284,7 @@ pub fn smart_chunk(content: &str, target_tokens: usize, overlap_pct: usize) -> V
             }
         };
 
+        let cut_offset = snap_to_char_boundary(content, cut_offset);
         let chunk_text = content[start_offset..cut_offset].trim().to_string();
         if !chunk_text.is_empty() {
             let heading = extract_heading(&chunk_text);

src/store.rs

@@ -63,8 +63,6 @@ CREATE TABLE IF NOT EXISTS files (
     docid        TEXT
 );
 
-CREATE INDEX IF NOT EXISTS idx_files_docid ON files(docid);
-
 CREATE TABLE IF NOT EXISTS chunks (
     id          INTEGER PRIMARY KEY,
     file_id     INTEGER NOT NULL REFERENCES files(id) ON DELETE CASCADE,
@@ -130,8 +128,11 @@ impl Store {
         };
         if !has_docid {
             self.conn
-                .execute_batch("ALTER TABLE files ADD COLUMN docid TEXT; CREATE INDEX IF NOT EXISTS idx_files_docid ON files(docid);")?;
+                .execute_batch("ALTER TABLE files ADD COLUMN docid TEXT;")?;
         }
+        // Always ensure the index exists (safe for both fresh and migrated DBs).
+        self.conn
+            .execute_batch("CREATE INDEX IF NOT EXISTS idx_files_docid ON files(docid);")?;
         Ok(())
     }