feat: Add Reader.RawValue and Reader.Offset for raw value capture

[keelerm84]

Port of the v3 RawValue/Offset work (#45) to the v4 module line.

RawValue returns the next JSON value's raw bytes as a zero-copy slice of the
input (cap == len), fully validated: scalars via the tokenizer, containers via a
string-aware boundary scan plus encoding/json validation of the captured span.
Offset returns the current byte position for capturing a value parsed in place.

v4 has no easyjson backend, so the port drops the build tags and the
easyjson-specific comments/placeholder that the v3 change carried; the logic is
otherwise identical.

---

Note

Medium Risk
New parsing surface on the core tokenizer with container boundary scanning; behavior is heavily tested and additive, but mistakes could affect JSON acceptance or slice aliasing guarantees.

Overview
Adds Reader.RawValue() and Reader.Offset() so callers can capture the next JSON value as a zero-copy slice of the input (cap == len), or slice the buffer around an in-place parse without RawValue’s extra container scan.

RawValue validates scalars via the existing RFC 8259 tokenizer; for arrays/objects it uses a string-aware delimiter scan and then encoding/json.Valid on the captured span. Unread-token paths (e.g. after Null() probes) are handled explicitly. A new errMsgInvalidValue backs invalid container content.

Offset reports the next consume position (or the start of a pushed-back token), documented for start/end span capture with JSON whitespace trimming.

Coverage includes unit tests, encoding/json differential/oracle tests, generative runs, and scalar grammar fuzzing.

^{Reviewed by Cursor Bugbot for commit c72dc8d. Bugbot is set up for automated code reviews on this repo. Configure here.}