build: prepare Stage 2 foundation

[edloidas]

Preparatory commit that establishes shared infrastructure for all Stage 2 features (#25). Must land before any individual Stage 2 feature work begins.

Token Allocation

Reorganize TokenType enum with all Stage 2 token IDs allocated upfront (31 total, semantically grouped). Existing tokens are renumbered for logical grouping:

Group	Tokens
Literals	NUMBER
Dice operators	DICE, DICE_PERCENT, DICE_FATE
Arithmetic	PLUS, MINUS, MULTIPLY, DIVIDE, MODULO, POWER
Comparison	GREATER, GREATER_EQUAL, LESS, LESS_EQUAL, EQUAL
Grouping	LPAREN, RPAREN, COMMA
Keep/drop	KEEP_HIGH, KEEP_LOW, DROP_HIGH, DROP_LOW
Explode	EXPLODE, EXPLODE_COMPOUND, EXPLODE_PENETRATING
Reroll	REROLL, REROLL_ONCE
Success	FAIL
Functions	FUNCTION
Keywords	VS
End	EOF

ComparePoint Type

Define canonical ComparePoint in src/types.ts:

type CompareOp = '>' | '>=' | '<' | '<=' | '=';
type ComparePoint = { operator: CompareOp; value: ASTNode };

value is ASTNode (not number) to match the existing pattern where DiceNode.count and DiceNode.sides are both ASTNode. This supports computed thresholds (>=ceil(5)) and negative compare values (=-1 for Fate dice) via unary minus.

Lexer: Full-Accumulation scanIdentifier

Rewrite scanIdentifier from incremental peek-based scanning to full-accumulation: collect all consecutive alpha chars, then classify against a keyword lookup table. Resolves the d-in-round problem, d% scanning, f vs floor disambiguation, and all future identifier keywords.

Special cases after accumulation:

• Bare d followed by % → DICE_PERCENT
• df → DICE_FATE

Operator Scanning

Add to the operator switch in nextToken():

• Comparison: >, >= (maximal munch), <, <= (maximal munch), =
• Explode: !, !! (maximal munch), !p/!P
• Comma: ,

Parser: ComparePoint Utilities

Add two methods to the Parser class for use by Stage 2 modifier parsers:

• isComparePointAhead(): boolean — peeks for comparison operator tokens
• parseComparePoint(): ComparePoint — consumes operator + value expression

Comparison tokens get BP = -1 (terminators) so they never interfere with the main Pratt loop.

Rationale

All 8 Stage 2 features were planned in isolation and independently assigned conflicting token IDs starting at 15, used 3 different naming conventions for comparison tokens, and defined ComparePoint in 4 incompatible ways. A consilium review identified these as critical integration blockers. This issue resolves all cross-cutting conflicts in a single upfront pass.

See .claude/docs/STAGE2.md for the full design document.

_{Drafted with AI assistance}