refactor(memo): fold encode_request and encode_name_note policy

[craftsoldier]

What

Extract memo::classify_action(action, ua) so encode_request and encode_name_note share one copy of the per-action ua policy check (Release must have empty ua; Claim and Update must have non-empty ua).

Why

Both encoders previously had a byte-identical 6-arm match action { Release if !ua.is_empty() => FieldCount, Claim | Update if ua.is_empty() => EmptyArg, ... } block. A future edit to one but not the other is a silent strictness divergence in the memo grammar. One call site, one audit, no drift. This is the audit observation D2 (encoder drift) in docs/code_review.md.

Verification

$ cargo test -p zns-verify
test result: ok. 9 passed  (memo_tests)
test result: ok. 5 passed  (vectors)         <- sacred, unchanged
test result: ok. 5 passed  (verify_tests)
test result: ok. 5 passed  (commitment_tests)
test result: ok. 1 passed  (doc-tests)

No new clippy or doc warnings introduced by this change (pre-existing type_complexity warnings on the public parse_*_memo signatures are unchanged).

What this does NOT do

• Does not touch any memo bytes, field ordering, grammar strictness, parser behavior, commitment code, or any cross-language test vector.
• Does not introduce or remove any public API entry point.
• The companion commitment-side refactor (inlining tagged_zns_hash's absorb steps) is tracked in a separate PR, since they target independent audit observations and can land or revert independently.