Address consumer feedback: serde-with for plain integers, schemars, ParseError, parse(), precision, Div

[lthiery]

Implements items 1–6 from downstream (conveyor) feedback. Every item has test coverage in bytescale.

1. #[serde(with = "humanbyte::serde")] on plain u64/usize

New humanbyte::serde module, generic over integer types (TryInto<u64>/TryFrom<u64>), plus humanbyte::serde::map_keys for BTreeMaps keyed by byte sizes. Human-readable formats (JSON/TOML) get "1.5 KiB" strings; binary formats get raw u64. This should let conveyor drop its hand-rolled byte_size.rs and consolidate off ubyte.

Note: humanbyte::serde was previously the re-export of the serde crate (used by derive-generated code). That re-export moved to a #[doc(hidden)] serde_crate alias. Anyone importing humanbyte::serde::Serialize directly (unlikely outside the derive) breaks.

2. schemars feature

impl JsonSchema emitted by the full HumanByte bundle when the feature is on (also available à la carte as HumanByteSchema). Schema is "type": ["string", "integer"], matching what the deserializer actually accepts. Targets schemars 1.x — say the word if conveyor is still on 0.8/0.9 and I'll widen it.

3. ParseError is now the real FromStr::Err

Derived FromStr returns humanbyte::ParseError (implements std::error::Error under std), so ?-chaining into TAF-style error types works. ParseError is no longer dead code.

4. humanbyte::parse(&str) -> Result<u64, ParseError>

The parse logic moved out of the generated code into the core crate (the derive now just calls it — less token bloat per derive, too). No more throwaway struct DiskSize(u64).

5. Configurable precision

humanbyte::to_string_with_precision(bytes, format, precision) + derived size.to_string_with_precision(Format::IEC, 2) → "1.00 TiB". to_string keeps its 1-decimal default, so no display output changes.

6. Div/DivAssign (+ Rem)

size / size → u64 (a dimensionless count: file_size / chunk_size chunks), size / scalar → size, DivAssign<scalar>, and Rem<size> since chunking is the motivating case. Also made Format Copy + PartialEq + Eq.

Minor items — deliberately not done

• Format::IEC → Format::Iec: upper_case_acronyms only fires at the definition site (allowed here), consumers don't lint on use — renaming would break all 7 existing use sites for no lint win.
• HumanByteRange removal: left in place; that's a separate decision.

Breaking changes (needs 0.3.0 at release)

• FromStr::Err: String → ParseError
• humanbyte::serde path repurposed
• Unit::from_str error type likewise ParseError