bug fixed in read_to_buffer

[jsulmont]

fixed a bug in buffer.jl triggered when read_to_buffer is called on a non empty buffer.
This is related to #32397.

[bicycle1885]

Looks good to me if you add tests.

[jsulmont]

Looks good to me if you add tests.

@bicycle1885 I've added a few tests

[jsulmont]

Hi @JeffBezanson @bicycle1885 -- any feedback? Am I missing anything?
As I said, this can only happens when read_to_buffer is called when buffer.size > 0, which implies:

1. it is called more than once, which means the size of the encoded input is greater than 512, and,
2. since encoded characters are consumed by 4 (multiple of 512), some input has been discarded (because not in the encoding table) causing read_to_buffer to be called again (possibly with buffer.size > 0).

IMO using code coverage might have revealed that the existing tests weren't triggering that part of the code.

[bicycle1885]

This looks good to me now But since I have no permission to merge, we need to ask someone else to merge.