Fix `Symbol#inspect` of UTF_16/UTF_32

[yui-knk]

Stop to append byte before inspect string.
For example, when an encoding of symbolBytes is UTF_16LE, "a" is
[0x61, 0x00]. If we append ":" (0x3A) to symbolBytes before
inspect it, bytes are [0x3A, 0x61, 0x00] with UTF_16LE encoding.
This is not what we want to get. This commit chnages the order of
inspecting and appending to avoid this.

Ref: https://github.com/ruby/ruby/blob/v2_5_0/string.c#L10402

[enebo]

@yui-knk You seem to even fix a second bug in here where we seem to be adding :" at the front of a symbol but add no closing ". I might change this code now that you have fixed this because we potentially make 3 instances of RubyString depending on the symbol being inspected. I think we can reduce this to just one.

[enebo]

Actually I will not be planning on changing this. 1) :sym.inspect is exceedingly rare in hot code 2) guts of bytelist vs RubyString and ability to determine CR_7BIT is much simpler if we make a string first. Working around that to defer making the string would involve some new code.

I did glance at MRI and they remove some of this cost by using memcopy/memmove and set the ':' and contents of the string. We could optimize in this way if we wanted but due to 1) above I am not inclined to put in that extra effort :)

[yui-knk]

ability to determine CR_7BIT is much simpler if we make a string first

I agree :)