Replace tree-sitter parser with nom

[toddATavail]

Summary

Replace the tree-sitter parser with a hand-written nom combinator parser, delivering a complete rewrite of the parsing layer while preserving the existing compiler backend (IR, optimizer, evaluator) byte-for-byte.

What changed

New nom parser (parser/combinators.rs, parser/errors.rs)

• Full nom combinator parser covering the entire xDy grammar: constants, standard dice (xDy), custom dice (xD[...]), arithmetic (+, -, *, /, %, ^), grouping, drop lowest/highest, formal parameters, environmental variables, and dynamic expressions.
• Correct operator precedence: exponentiation binds tighter than unary negation (-a^2 is -(a^2)), matching tree-sitter behavior.
• Negative constant handling: -2147483648 (i32::MIN) parses correctly via a dedicated negative_overflowed_constant combinator that handles the sign+digits boundary as a unit.
• Constant overflow saturates to i32::MAX/i32::MIN instead of failing.
• Identifiers accept . and inline whitespace, matching the tree-sitter grammar ([\p{L}_][\p{L}\p{N}\p{Z}._-]*).
• Structured error propagation via cut at commit points (after binary operators, delimiters, d/D operator, drop keyword), so errors propagate instead of being silently swallowed.

Typed AST (ast.rs)

• Strongly typed AST with Function, Expression, DiceExpression, ArithmeticExpression, DropDirection, and Span types.
• ASTVisitor trait with accept() dispatch on all expression types, enabling clean tree-walking passes.

AST-walking compiler (compiler.rs)

• Compiler walks the nom AST and emits byte-identical IR to the old tree-sitter compiler.
• Exposed as public API for power users who want to drive the parse → AST → codegen pipeline manually.
• All lifetime parameters renamed from 'a to 'src (and 's/'e/'r to 'src/'env/'rng) for self-documenting code.

Diagnostics module (diagnostics.rs)

• Two-layer error reporting: enhanced parser error propagation + fix-and-retry doctor.
• 11 diagnostic kinds: UnclosedDelimiter, MissingRightOperand, MissingLeftOperand, MissingDiceFaces, IncompleteDropClause, BareIdentifier, IncompleteParameterDefinition, TrailingInput, EmptyExpression, UnexpectedToken, UnexpectedEof.
• Structured Diagnostic type with SourceSpan, Suggestion (including complete corrected source), and Placeholder regions with valid_kinds for UI integration.
• OffsetMap tracks position adjustments across fix-and-retry iterations.
• Bare identifier detection handles d/D splitting (xD6 → {x}D6 or {xD6}) and expression-context variables (4 + hello → 4 + {hello}).
• Missing operand diagnostics offer both insert-placeholder and drop-operator suggestions.

S-expression support (s_expr.rs)

• S-expression writer for AST pretty-printing (debugging, golden-file testing).
• S-expression reader (read_s_expr()) that roundtrips perfectly with the writer.
• Variables and parameters containing whitespace are quoted with double quotes in both directions.

Comprehensive test infrastructure

• 3,241 lines of parser tests (tests/parser.rs) covering all grammar constructs, edge cases, Unicode, whitespace tolerance, and operator precedence.
• 93 data-driven parse cases (tests/test_parse.txt) covering all expressions from test_compile_unoptimized.txt plus overflow boundaries, Unicode, embedded spaces, and whitespace tolerance.
• 33 data-driven error diagnostic cases (tests/test_parser_errors.txt) with exact match on diagnostic kind, span, message, suggestions, and placeholders.
• 345 lines of diagnostics tests (tests/diagnostics.rs) including corrected source validity and performance (2000 diagnose() calls < 1s).
• Tests organized into submodules: compiler, diagnostics, evaluator, histogram, optimizer, parser.

tree-sitter decoupling

• Removed tree-sitter, tree-sitter-xdy, and cc dependencies from the xdy crate.
• Deleted the old tree-sitter Compiler struct, SyntaxTreeVisitor trait, and CanVisitSyntaxTree trait (~800 lines).
• tree-sitter-xdy crate preserved untouched as a standalone package.

Documentation overhaul

• Root README: project brochure (language features, examples, performance, planned work).
• Crate README: technical deep-dive covering compilation pipeline, ISA reference, VM model, optimizer internals, histogram engine, and diagnostics.
• Railroad diagram generation via ebnsf: EBNF grammar in xdy/grammar/xdy.ebnf compiled to SVG at build time and inlined into parser Rustdoc.
• Mermaid diagrams via aquamarine in compiler, optimizer, and evaluator Rustdoc.
• Rustdoc # Lifetimes sections on all major types.

Dependency upgrades

• rand 0.8 → 0.10 (thread_rng() → rand::rng(), gen_range() → random_range()).
• criterion 0.5 → 0.8.

Build & tooling

• Top-level justfile with verify, fmt, clippy, test, bench, doc recipes.
• CompilationError now carries ParseError with lifetime parameter for full diagnostic fidelity.
• EvaluationError updated correspondingly. Both lose Copy (ParseError contains Vec).

Performance

Benchmarked across 316 expressions: nom delivers 48% aggregate speedup (median 62% per expression), with 99% of cases faster than tree-sitter.

Compatibility

• All existing end-to-end test cases pass unchanged.
• Public API (compile, compile_unoptimized, evaluate) preserved with the same signatures (modulo lifetime on CompilationError).
• IR, optimizer, and evaluator are completely untouched.

Test plan

• All existing end-to-end tests pass (tests/*.txt)
• New parser tests cover all grammar constructs
• Diagnostics tests verify error recovery and suggestions
• S-expression roundtrip tests
• Performance benchmarks confirm speedup
• cargo clippy clean
• cargo test all pass