fix(ssz): correct JustificationValidators LIMIT formula in rule doc

[tcoratger]

Summary

The SSZ-patterns rule doc showed LIMIT = HISTORICAL_ROOTS_LIMIT * HISTORICAL_ROOTS_LIMIT for the JustificationValidators example, but
the type stores one bit per (tracked root, registered validator)
pair. The second factor must be the validator registry limit. The
code (state/types.py:167) already uses the correct product — only
the rule doc was wrong.

Why it matters

BaseBitlist.LIMIT directly controls merkleization depth:

Source	Formula	Value (bits)	Chunk count	Tree depth
Code	2^18 * 2^12	2^30	2^22	22
Rule doc (was)	2^18 * 2^18	2^36	2^28	28

A second-client implementer treating .claude/rules/ssz-patterns.md
as canonical would compute a different hash_tree_root for the same
state field. Different state roots → different block roots → silent
consensus divergence on the first network exchange. Exactly the kind
of doc/code drift this rule file is meant to prevent.

Changes

• .claude/rules/ssz-patterns.md — fix the formula in the example
  and add an inline rationale block explaining why the product matters
  (consensus-critical merkle depth). Also define VALIDATOR_REGISTRY_LIMIT
  alongside HISTORICAL_ROOTS_LIMIT so the math is self-contained.
• .claude/rules/ssz-patterns.md — update the frontmatter path
  glob and the architecture diagram to reflect that containers moved
  to forks/<fork>/containers/ in the recent refactor.
• state/types.py — expand the one-line docstring on
  JustificationValidators to document the layout and the
  consensus-critical nature of the LIMIT product. One sentence per
  line per the documentation rules.

Test plan

• ruff check, ruff format --check, ty check pass
• 104/104 tests in tests/lean_spec/forks/lstar/state and
  tests/lean_spec/forks/lstar/forkchoice pass
• No code-level value changed — the code was already correct, only
  the doc was wrong

🤖 Generated with Claude Code