Short-circuit einsum pre-cache symmetry/identity work before path lookup

[spMohanty]

whest._einsum already has a global LRU path cache keyed by (subscripts, shapes, optimize, symmetry, identity, inner_sym) and used by einsum and einsum_path.

Observed overhead still occurs before cache hit on repeated calls:

• subscripts split/parse work,
• _symmetry_fingerprint(...) materialization,
• _identity_pattern(...) grouping,
• symmetry reconstruction that only affects cache keying.

Target optimization opportunities:

• add quick no-symmetry/no-repeated-object fast paths,
• avoid constructing full symmetry fingerprints when no SymmetricTensor operands are present,
• cache/fuse per-op symmetry metadata where it is reused across repeated calls.

Files

• src/whest/_einsum.py

Acceptance Criteria

• Do not change cache-hit correctness or path result.
• Keep einsum/einsum_path behavior unchanged.
• Reduce pre-cache call overhead for repeated einsum patterns.