codegen: route bytes_fields paths through ::buffa::bytes:: re-export

[iainmcgin]

Part of #76.

Generated code referenced ::bytes::Bytes directly at 12 sites, all gated on the bytes_fields config option. That requires every consumer enabling bytes_fields to add bytes as a direct Cargo dependency, even though buffa re-exports it (pub use ::bytes;). This routes those sites through ::buffa::bytes::Bytes instead — same convention as ::buffa::bytes::BufMut, ::buffa::alloc::*, and ::buffa::__private::* elsewhere in emitted code.

Mechanical s/::bytes::Bytes/::buffa::bytes::Bytes/ across impl_message.rs (5), impl_text.rs (3), message.rs (2), oneof.rs (1 + 1 doc-comment), view.rs (1). The codegen_integration snapshot assertion that spanned the Bytes::from( open-paren was loosened since the longer path now triggers prettyplease line-wrap there. WKT regen only touches Any (the only WKT with a bytes_fields entry).

The unconditional case (#74's to_owned_from_source signature) is independent of this and lands separately; this PR has a trivial 1-line overlap in view.rs:61 that whichever merges second will absorb.

Audit also confirmed: ::serde:: / ::serde_json:: / ::arbitrary:: direct paths are by design (documented consumer-side deps for generate_json / generate_arbitrary); ::alloc:: / ::std:: / ::hashbrown:: / ::once_cell:: have zero non-prefixed emissions.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}