This should reduce the overhead of inspecting by avoiding the
intermediate StringBuilder/char[] and subsequent conversion to
byte[]. There are a few extra objects created along the way
(java.lang.String.getBytes, RubyString for class name, etc) but
overall memory use should reduce around 3x: a US-ASCII
string would need 2x byte size in the StringBuilder, and the
builder is still alive when we create the eventual byte[] for a
total of 3x max memory use during inspect.

Relates to #4127, but does not constitute a fix for memory issues
there.

Inspect directly into a RubyString and avoid StringBuilder/char[].