Define shared LaTeX tokens and text/math mode transitions #296

CSharpMath.Atom.LaTeXParser and CSharpMath.Rendering.Text.TextLaTeXParser currently scan control sequences, braces, whitespace, and math delimiters independently. Define the internal lexical and parser-mode contract that both can consume before attempting a behavioral migration.

Scope:

• Define shared token kinds with source spans for control words/symbols, groups, whitespace, raw text, $, and $$ boundaries.
• Define explicit math/text mode transitions and a stack discipline for nested groups.
• Preserve current public parser and painter entry points; this issue is an internal architecture seam, not a package merge.
• Do not bundle the global-settings redesign into this work.

Acceptance criteria:

• Shared token/mode-transition code is consumed by focused tests for both parser contexts.
• Escaped dollars, braces, control-word termination, nested groups, and math-to-text-to-math transitions retain exact source positions.
• Malformed or unmatched delimiters produce deterministic errors without leaking parser state.
• Existing math and text parser suites remain green with no public API break.

Parent epic: #58.