view: zero-copy Bytes in to_owned via slice_ref of source buffer

[iainmcgin]

Fixes #52.

Problem

PR #51 backed Any.value (and any field in bytes_fields) with bytes::Bytes so cloning is a refcount bump. But the view→owned path — which production decode pipelines typically hit — was still emitting Bytes::copy_from_slice(self.value), allocating and copying just like the old .to_vec().

Approach

Rather than threading a &'a Bytes through decode_view (which doesn't compose with OwnedView<V> — the Bytes handle moves into self after decode, so a stored reference would dangle), this passes the source buffer at conversion time:

• New MessageView::to_owned_from_source(&self, source: Option<&Bytes>) -> Self::Owned, with a default impl delegating to to_owned_message (so hand-written impls are unaffected).
• New runtime helper buffa::view::bytes_from_source(Option<&Bytes>, &[u8]) -> Bytes: returns Bytes::slice_ref when the slice lies within the source (pointer-range check mirroring slice_ref's precondition), Bytes::new() for empty, else copy_from_slice. Slices outside the source (e.g. on a manually-constructed view) fall back to copy rather than panic.
• Generated to_owned_message() now calls to_owned_from_source(None); the new method body emits bytes_from_source(__buffa_src, ...) for bytes_fields and recurses into nested/repeated/oneof/map message values with to_owned_from_source(__buffa_src).
• OwnedView::to_owned_message() now calls to_owned_from_source(Some(&self.bytes)), so the 'static wrapper gets zero-copy automatically.

No decode-path changes, no view-struct field additions.

Tests

• 5 unit tests for bytes_from_source (None, in-range, out-of-range, empty, full-range).
• End-to-end zero-copy assertions (pointer-range checks, not just byte equality) for Any via both the direct trait path and OwnedView.
• buffa-test coverage for repeated/optional/oneof/nested-message bytes shapes under use_bytes_type().

Out of scope / follow-ups

• map<K, bytes> values are still Vec<u8> regardless of bytes_fields (pre-existing; map_to_owned_expr hard-codes .to_vec()).
• The issue's string_fields suggestion is left for later.

[github-actions]

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