Defer UTF-8 validation in struct deserialization

[alexrutar]

Struct deserialization should be improved to reduce the number of UTF-8 conversion checks.

• Convert Read::identifier into an identifier_bytes method.
• Validate string, comment, and preamble directly from the bytes using to_ascii_lowercase
• Otherwise, perform UTF-8 validation (skipping if input is str).
• Expose the raw bytes to any Deserialize impl so that if deserializing fields into a struct, the struct names can be compared against the raw bytes directly.
• Implement Deserialize in an example or in the entry module. Since all standard biblatex entry keys fields are ascii and normalized to lowercase, comparisons can be done directly from bytes using to_ascii_lowercase.