[RobertoPrevato]

• Improve header ergonomics: dictionary-style assignment and Headers.update()
  accept strings or bytes, so response.headers["X-Example"] = "value" is
  supported while headers remain stored internally as bytes. String names use
  ASCII and string values use Latin-1; unrepresentable characters raise
  UnicodeEncodeError. Assignment still replaces all case-insensitive matches.
• Add Headers.get_text_values(name, encoding="latin-1") to decode matching
  values on demand as a tuple, preserving repeated header entries.
  get_first_text() returns the first value as a string, or None when absent;
  get_single_text() returns a string and requires exactly one matching entry,
  raising ValueError otherwise. Both also accept an explicit encoding.
  Existing byte-valued reads and low-level byte-based methods remain unchanged.
• Complete header dictionary symmetry: lookup, membership, and deletion accept
  string or byte names; lookup always returns byte values. Add Headers.add_text()
  to append without replacing matches, and Headers.from_values() to build a
  byte-backed collection from mappings or iterables while preserving duplicates.
• Use the same ASCII-name/Latin-1-value policy for HTTP client headers= and
  default_headers=, route header filters, and testing-client iterable headers.
  Previously, HTTP client string values required ASCII. Testing-client dictionaries
  retain their historical UTF-8 encoding for backward compatibility.
  Pass explicitly encoded bytes for field-specific encodings.
• Preserve raw header values when sending HTTP/2 requests, without decoding them
  as UTF-8, so Latin-1 values and explicitly encoded bytes reach the wire unchanged.