feat(math): add context-free language operations (#1894)

[morluto]

Summary

Implements the domain with 3 atomic operations for context-free grammar analysis.

Closes #1894

Operations

Operation	Description
grammar.symbol_profiles.compute	Compute nullable nonterminals via fixed-point iteration
grammar.dependency_graph.compute	Compute the dependency graph (A -> B if A has a rule containing B)
grammar.first_sets.compute	Compute FIRST sets via fixed-point iteration

Design

• Fixed-point iteration: Nullable symbols and FIRST sets are computed via standard fixed-point iteration until convergence.
• Exact combinatorial analysis: All operations use exact set operations over bounded grammars (≤32 nonterminals, ≤256 rules).
• Dependency graph: Built directly from rule bodies, identifying which nonterminals appear in the body of which rules.

Continue this on Linzumi

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[morluto]

Deep review summary

Verdict: REQUEST CHANGES — FIRST sets are wrong across nullable prefixes, and malformed grammar symbols can be treated as nullable.

1. first_sets.compute stops after the first nonterminal unconditionally

For a production S → A B, FIRST must continue into B whenever A is nullable. The implementation adds FIRST(A) and then always breaks. The branch intended to inspect nullability is unreachable because symbol in first is necessarily true for every recognized nonterminal.

Concrete counterexample:

S → A B
A → ε
B → b

The correct result is FIRST(S) = {b}. The implementation returns the empty set. More generally, if FIRST(A)={a,ε}, then FIRST(S) must include both a and FIRST(B)\{ε}.

Compute nullable first, then scan each production left-to-right: union the non-epsilon FIRST set of each nonterminal, continue only while it is nullable, stop at a terminal or nonnullable symbol, and represent whether the whole body derives epsilon explicitly.

2. Undeclared body symbols can make a head nullable

The grammar validator checks production heads but never requires every body symbol to belong to exactly one of the terminal or nonterminal alphabets. In compute_symbol_profiles, an unknown symbol is neither a terminal nor a known nonterminal, so the loop leaves all_nullable=True; a rule such as A → UNKNOWN incorrectly marks A nullable.

Require unique, disjoint terminal/nonterminal alphabets; every body symbol must be declared; start/head symbols must be canonical; and duplicate productions should be rejected or canonicalized. If a symbol appears in both alphabets, the current algorithms silently disagree about its role.

3. Result identity is positional and the new _ops domain duplicates an owner

FIRST/nullable tuples carry no nonterminal labels and depend on caller ordering. Return structured (nonterminal, value) rows or require canonical ordering. This should also extend the existing context-free-language owner rather than creating a parallel context_free_languages_ops namespace with overlapping semantics.

The dependency graph construction is correct once the grammar value is well formed.