Add a unit argument to the length validator (utf16 / bytes)

[ChrisJr404]

Closes #250.

Motivation

The length validator counts strings in Unicode scalar values (chars, via str::chars().count()). That doesn't match how an HTML form field's maxlength or JavaScript's String.length count characters — both use UTF-16 code units. So for non-ASCII input the backend and the frontend can disagree about whether a value is valid.

Concretely, "🖐🏽" is 2 chars, but 4 UTF-16 code units and 8 UTF-8 bytes. A form that lets the user type up to 4 "characters" of that emoji can produce a value the current length validator rejects (or accepts) inconsistently with the browser.

In #250 you suggested handling this with a mode/unit parameter on the length validator rather than a separate length_utf16 validator, so that unicode / utf-16 / bytes are all covered by one option. This implements that.

What this does

Adds an optional unit argument to length:

#[validate(length(max = 10, unit = "utf16"))]
title: String,

#[validate(length(min = 1, max = 255, unit = "bytes"))]
slug: String,

• unit = "chars" (default) — Unicode scalar values, i.e. today's behaviour
• unit = "utf16" — UTF-16 code units (matches maxlength / JS String.length)
• unit = "bytes" — length of the UTF-8 encoding

The unit only changes how string-like values are measured. Collections (Vec, HashMap, arrays, …) are still measured by their number of elements, and the unit is ignored for them. When a unit is set it's also added to the error as a unit param, alongside min/max/equal.

Implementation notes

• ValidateLength gains validate_length_with_unit and length_with_unit, both with default implementations. length_with_unit defaults to length(), so every existing impl (including user-provided ones that only implement length()) keeps working unchanged. The string impls and the deref/Cow/Option forwarders override it so the unit is honoured through wrappers like &String and Option<String>.
• validate_length now delegates to validate_length_with_unit(LengthUnit::Chars, …), so its behaviour is identical to before. LengthUnit is exported from the crate root.
• The derive parses unit = "chars" | "utf16" | "bytes" (a couple of aliases like unicode/utf8 are accepted too) and emits the unit-aware call only when the argument is present; otherwise it emits the exact same validate_length(...) call as before, so output for existing code is unchanged.

Tests

Added unit tests for the trait (utf16/bytes/chars, ASCII equivalence, forwarding through &String/Option/Cow, and the collection no-op) and derive integration tests in validator_derive_tests/tests/length.rs (utf16 max, bytes min/max, explicit-chars-equals-default, the exposed unit param, and Option<String>).

cargo test -p validator and cargo test -p validator_derive_tests --test length pass; cargo fmt --check and cargo clippy are clean. Also updated the README length section and the changelog.