Support chained equalities in the answer and response

[m-messer]

Problem

Chained equalities were silently truncated. parse_expression split on = but
only used the first two parts, so everything after the second operand was
discarded — never parsed, never validated, no error. x = y = z became
Eq(x, y), and a wrong third term was marked correct:

Response	Answer	Before	After
x = y = w	x = y = z	✅ correct	❌ incorrect
x = y = 5	x = y = 7	✅ correct	❌ incorrect
x = y = z	x = y	✅ correct	❌ incorrect

The preview showed the student's full x=y=z back to them, so nothing on screen
hinted that only x = y was being graded.

Changes

• parse_expression now parses every operand and returns the conjunction of the
  equalities between adjacent ones: x = y = z → And(Eq(x, y), Eq(y, z)). A
  single equality is still a bare Equality, so that path is unchanged. An empty
  operand (x = = y) is now a parse error.
• check_equality_equivalence in app/context/symbolic.py, mirroring
  check_inequality_equivalence. Chains are compared by solution set, since a
  chain carries no meaning in the order it is written — z = y = x states what
  x = y = z states, and scaled or offset chains (2x = 2y = 2z) match too.
  Non-affine chains fall back to matching parts one-to-one.
• New CHAIN_NOT_EQUALITY / EQUALITY_NOT_CHAIN feedback tags for a chain
  compared against a single equality.
• Docs and tests: 31 new tests, 1809 passed, 2 skipped.

Stacked on #276

Based on feature/inequalities, not main. Merge #276 first, then retarget
this to main; until then the diff against main includes #276's commits.

🤖 Generated with Claude Code