assertInteger: can not decode float as int for valid input from chunked reader

[mrVanboy]

Hi,
we found a bug during unmarshaling large responses from HTTP.Request.Body. The next conditions should be met:

• Iterator reads data from io.Reader,
• expected value is 0 (zero)
• Iterator.tail == Iterator.head + 1
• Iterator.tail < len(Iterator.buf)
• value in the buffer after Iterator.tail is presented from the previous read and has '.' character.

Basically, let's assume that Iterator reads data from the reader which returns less data, than the size of the buffer. After a couple of reads, the buffer will have the content like: 0.0, 0.0, 0.
The first item is from the latest read, other items are from the older readers, so have:

buf: []bytes(`0.0, 0.0, 0`)
head: 0 // ---^
tail: 1 // ----^

When the Iterator.ReadInt is called, it reads zero and checks if the value is an integer by calling assertInteger() method, which currently looks like:

func (iter *Iterator) assertInteger() {
	if iter.head < len(iter.buf) && iter.buf[iter.head] == '.' {
		iter.ReportError("assertInteger", "can not decode float as int")
	}
}

The only issue is, that it checks for iter.head < len(iter.buf) instead of iter.head < iter.tail and the assertion is done against the dot . from the previous read.

The test for replicating this behavior will be provided in ongoing PR.

[akovalyov]

@AllenX2018 any chance this will get into the next tag?