stdlib: add grapheme traversal and locale-independent case mapping

Context

silk/unicode now ships Unicode 17.0.0 NFC and NFD normalization. This issue tracks the
remaining post-stable Unicode functionality: extended grapheme-cluster traversal and full
locale-independent default upper/lower case mapping.

Locale-sensitive collation remains tracked separately in #162.

Current behavior

import silk.unicode { normalizeNfc, normalizeNfd }

Normalization is available and ordinary string equality remains exact. There is currently no
grapheme cursor and no upper- or lower-case transformation API.

Intended surface

pub fn nextGrapheme(
  value: string,
  cursor: GraphemeCursor
) -> Option<GraphemeStep>

pub effect fn toLowercase(
  value: string
) -> String ! OutOfMemoryError ? &mut Allocator

pub effect fn toUppercase(
  value: string
) -> String ! OutOfMemoryError ? &mut Allocator

GraphemeStep reports byte start/end offsets and the next cursor. It does not retain a borrowed
string view.

Requirements

1. Grapheme traversal implements Unicode 17.0.0 extended grapheme clusters from UAX #29.
2. Traversal is allocation-free and reports UTF-8 byte boundaries.
3. Case conversion implements full locale-independent default lower/upper mappings, including
   contextual mappings.
4. Case conversion allocates an owned String and exposes OutOfMemoryError plus Allocator.
5. Language tailoring, case folding, title casing, and locale-sensitive collation are out of scope.
6. Tables remain generated ordinary Silk data with no compiler privilege.
7. Ordinary string equality remains exact and normalization-free.

Acceptance criteria

• A base scalar and combining mark remain one grapheme.
• Emoji ZWJ and regional-indicator conformance cases match Unicode 17.0.0 data.
• Traversal returns exact UTF-8 byte offsets without allocating.
• Lower/upper conversion covers a mapping that changes byte length.
• Contextual default mappings are covered.
• Evaluator, native LLVM, and direct Wasm agree.
• Generation is deterministic and the compiler contains no Unicode-policy special case.

Start with the silk/unicode implementation and the existing normalizeNfc and normalizeNfd surfaces. Review the proposed nextGrapheme, toLowercase, and toUppercase behavior against Unicode 17.0.0 and UAX #29, then verify agreement across the evaluator, native LLVM, and direct Wasm. Done means the listed grapheme, mapping, allocation, generation, and exact-equality criteria pass.