Accept Python bytes for raw binary payloads

[p1c2u]

Accept Python bytes for opaque binary payloads across OAS 3.0, 3.1, and 3.2,
implementing the binary-bytes validation design.

Binary detection now lives in one place (_binary.py) as pure, per-version
predicates, replacing the old inline format: binary branch. bytes acceptance
is confined to the type keyword wrapper; the "string" type checker is never
broadened.

Changes

• 3.0: type: string, format: binary accepts bytes (now also enforces
  maxLength/minLength by octet count). OAS30StrictValidator still rejects.
• 3.1/3.2 (default): canonical typeless schema (optionally a non-text
  contentMediaType, no contentEncoding) accepts bytes; type: string +
  non-text contentMediaType accepted as a documented pragmatic extension.
• New OAS31StrictValidator / OAS32StrictValidator: preserve JSON Schema
  string typing — accept typeless raw binary but reject bytes whenever
  type: string is asserted. validator_for still resolves dialect ids to the
  default validators.
• maxLength/minLength enforce octet length for raw bytes; enum/const
  stay active; pattern short-circuits bytes with no TypeError.
• Broadened encoded-text exclusion (any real contentEncoding, with no-op
  identity/binary/7bit/8bit allowlist); media-type parameters stripped
  and matched case-insensitively.

Migration

In 3.1/3.2, format: binary is no longer a binary marker (under 2020-12
format is an annotation). bytes validated against a 3.1/3.2
{"type": "string", "format": "binary"} schema is now rejected — model raw
binary with contentMediaType (ideally a typeless schema).