fix: support bytea in preferQueryMode=simple

[vlsi]

This adds text encoding for bytea.

Interestingly, the tricks with private final char[] hexDigits do not improve performance, and we probably should stick with Character.forDigit(...) for int -> hex conversions.

---

Unfortunately, we have two different handling of "stream underflow" with StreamWrapper and ByteStreamWriter.

Both of them might report their lengths, and both of them might happen to underflow or overflow.

PGStream handles StreamWrapper as follows (see org.postgresql.core.PGStream#sendStream):

• If there's not enough data in the stream => raise "Premature end of input stream"
• If there's too much data => just ignore it

PGStream handles ByteStreamWriter as follows (see org.postgresql.core.PGStream#send(ByteStreamWriter)):

• If there's not enough data in the stream => append zeros
• If there's too much data => throw an error

Note: ByteStreamWriter works in a push manner, so we can't easily stop it, so "throwing an error" on overflow might be ok.

However, it is unfortunate we have different behaviours regarding "pad with zeros" vs "throw on underflow". I think we should raise an error on underflow in both cases.

[vlsi]

I think this is ready, and I'm going to merge it after some time

[bokken]

Did you compare going directly to a byte[] with the ascii bytes rather than using a StringBuilder?

[vlsi]

The existing code was building SQL with StringBuilder, so I benchmarked SB only