Improve `JsonLocation` source handling: do not assume UTF-8 for byte sources

[cowtowncoder]

Although 2.9 improves handling of source references, and works nicely for JSON input, handling of byte sources (like byte[]) is problematic for binary formats like Smile and CBOR as it tries to decode a snippet in UTF-8. This won't work and instead we should either:

1. Display snippet in hex instead, OR
2. Just not display anything

Determination for this could be done in multiple ways; for example:

1. Use heuristics to see if first N bytes are UTF-8 safe
2. Use existing (or, possibly added) calls to determine if format is textual/binary

Either way, it'd be good to display UTF-8 contents for textual formats; and avoid problems for binary formats.

[cowtowncoder]

I think this is covered by #694 wrt binary formats, and perhaps by #658 (for textual).
Closing; should re-file for remaining issues, if any.