Parse async functions and await expressions

Parent: #439

## Summary

Add source-level `async` and `await` syntax without performing coroutine lowering.

## Scope

- Reserve `async` and `await` in the lexer with stable token names and spans.
- Parse `async fun name(...) -> T { ... }` wherever an ordinary function declaration is valid.
- Parse prefix `await expression` with documented precedence.
- Preserve async declarations and await expressions in the source AST.
- Support visibility parsing as `pub async fun`; continue rejecting every public `main` declaration.
- Produce focused diagnostics for missing `fun`, missing await operand, invalid declaration order, and malformed bodies.
- Keep `extern` and `export` async forms syntactically representable only if needed for a precise semantic rejection; do not silently accept them as synchronous functions.

## Completion criteria

- [ ] Lexer and parser unit tests cover valid nested expressions and malformed syntax.
- [ ] AST debug output exposes async declarations and await nodes.
- [ ] Existing identifiers and declaration parsing regressions remain green apart from the intentional keyword reservation.
- [ ] Parser changes do not add LLVM or runtime dependencies.

Semantic legality and types are handled separately from this issue.

Start by locating the lexer, parser, existing function-declaration and expression parsing, and AST debug output; read the related unit tests first. Done means async and await syntax is preserved in the AST, focused malformed-syntax diagnostics are covered, and existing declaration and identifier tests remain green without LLVM or runtime changes.

Written by the indexing model from the issue text.