feat: add percentile dice (d%)

[edloidas]

Add d% as an alias for d100. The notation d% and 2d% should produce DiceNode with sides = Literal(100).

Currently d% throws a parse error because % after d is scanned as DICE + MODULO. With the Stage 2 foundation (#34), the lexer already emits DICE_PERCENT for d%. This issue covers the parser and test work.

Parser Changes

• NUD (prefix d%): create DiceNode(Literal(1), Literal(100))
• LED (infix 2d%): create DiceNode(left, Literal(100))
• getLeftBp: DICE_PERCENT → BP.DICE_LEFT (40)
• No getRightBp needed — sides are always 100, no right-hand expression to parse

No new AST node type. Reuses DiceNode with synthetic sides = Literal(100).

Evaluator Changes

None. DiceNode with sides = 100 is already handled correctly by evalDice. Critical detection (result === 100 && 100 > 1) and fumble detection (result === 1) work naturally.

Expression/Rendered Output

• expression: shows 1d100 (canonical form)
• notation: preserves d% (original input)
• rendered: 1d100[42] = 42

Edge Cases

Expression	Expected
d%	1d100
2d%	2d100
D%	1d100 (case-insensitive)
d%+5	1d100 + 5
2d%kh1	Roll 2d100, keep highest
(2)d%	2d100 (computed count)
d % 3	Error — whitespace breaks the d% token, producing DICE + MODULO
d%%	Error — d% consumed as DICE_PERCENT, then bare % is MODULO with no NUD
10%3	Modulo (unchanged) — % after digit is still MODULO in the operator switch

Test Plan

• Parser tests for AST structure: d%, 2d%, d%+5, 2d%kh1, (2)d%
• Evaluator tests with MockRNG: verify sides = 100, total in [1, 100]
• Integration tests via roll(): d%, 2d%, d%+5
• Property test: Nd% total is in [N, N*100]
• Error tests: d%%, standalone d% followed by invalid token

_{Drafted with AI assistance}