refactor: remove duplicated code & split constraint solver

[desertthunder]

This PR refactors duplicated constraint, AST, source-position, and annotation helper logic into shared modules while preserving existing analyzer and LSP behavior.

Changes

Constraint solver modularization

• Split the constraint crate's solver implementation into focused modules for models, generation context, control-flow tracking, pattern compatibility, equality, calls, attributes, protocols, pattern constraints, joins, narrowing, state, and error recovery.
• Kept constraint types crate-local
• Deduplicated shared pattern/class compatibility helpers into and reused them across pattern validation and solver paths.

Shared parser and AST helpers

• Added parser-level SourceRange and byte-offset helpers so analyzer, constraints, and server code no longer need local line/column conversion copies.
• Added shared AstNode helpers for source ranges, statement/body traversal, child expression traversal, binding names, and target display.
• Added structured comprehension targets via Comprehension::target_node, allowing destructuring bindings to use AST-backed extraction instead of string splitting when available.
• Added parser test builders for common AST node shapes and coverage for source ranges, traversal helpers, comprehension targets, pattern binding names, and UTF-8 source-position conversion.

Analyzer, server, and core cleanup

• Updated analyzer walker/data-flow code to consume parser-owned source and binding helpers.
• Updated server diagnostics, inlay hints, semantic tokens, workspace parsing, and analysis code to use shared AST/source helpers.