fix: 3 SQL parser bugs — scientific notation, missing window functions, frame end default

- Lexer now handles scientific notation (1e10, 1.5e-3, 3E+5) in number tokens
- isFunctionKeyword() was missing NTILE, FIRST_VALUE, LAST_VALUE, NTH_VALUE,
  PERCENT_RANK, CUME_DIST — they were parsed as column refs instead of functions
- Window frame without AND (e.g. ROWS 5 PRECEDING) now defaults end to CURRENT ROW
  per SQL standard, instead of leaving it undefined

Author: teamchong

src/sql/lexer.ts

@@ -135,13 +135,18 @@ export function tokenize(sql: string): Token[] {
       continue;
     }
 
-    // Numbers
+    // Numbers (supports decimals and scientific notation: 1e10, 1.5e-3, 3E+5)
     if (/[0-9]/.test(ch)) {
       while (pos < len && /[0-9]/.test(sql[pos])) pos++;
       if (pos < len && sql[pos] === ".") {
         pos++;
         while (pos < len && /[0-9]/.test(sql[pos])) pos++;
       }
+      if (pos < len && (sql[pos] === "e" || sql[pos] === "E")) {
+        pos++;
+        if (pos < len && (sql[pos] === "+" || sql[pos] === "-")) pos++;
+        while (pos < len && /[0-9]/.test(sql[pos])) pos++;
+      }
       tokens.push({ type: TokenType.NUMBER, lexeme: sql.slice(start, pos), position: start });
       continue;
     }

src/sql/parser.ts

@@ -734,7 +734,7 @@ class Parser {
       const frameType = this.match(TokenType.ROWS) ? "rows" as const : (this.advance(), "range" as const);
       this.match(TokenType.BETWEEN); // consume optional BETWEEN keyword
       const start = this.parseFrameBound();
-      let end: NonNullable<SqlWindowSpec["frame"]>["end"];
+      let end: NonNullable<SqlWindowSpec["frame"]>["end"] = { type: "current" as const };
       if (this.match(TokenType.AND)) {
         end = this.parseFrameBound();
       }
@@ -793,7 +793,11 @@ function isFunctionKeyword(type: TokenType): boolean {
   return type === TokenType.COUNT || type === TokenType.SUM ||
     type === TokenType.AVG || type === TokenType.MIN || type === TokenType.MAX ||
     type === TokenType.ROW_NUMBER || type === TokenType.RANK ||
-    type === TokenType.DENSE_RANK || type === TokenType.LAG || type === TokenType.LEAD;
+    type === TokenType.DENSE_RANK || type === TokenType.NTILE ||
+    type === TokenType.LAG || type === TokenType.LEAD ||
+    type === TokenType.FIRST_VALUE || type === TokenType.LAST_VALUE ||
+    type === TokenType.NTH_VALUE || type === TokenType.PERCENT_RANK ||
+    type === TokenType.CUME_DIST;
 }
 
 function isKeywordIdentifier(type: TokenType): boolean {