Fix GH-23094: Use byte offsets in NumberFormatter parsing - #23318

The issue is: NumberFormatter rejects offsets inside UTF-8 sequences
For example:

<?php

$formatter = new NumberFormatter('en_US', NumberFormatter::DECIMAL);

$offset = 1; // Inside the two-byte UTF-8 encoding of "é".
var_dump($formatter->parse(
    "\u{00E9}123",
    NumberFormatter::TYPE_INT32,
    $offset
));
var_dump($offset);

?>

should return

bool(false)
int(1)

But in this PR, it is

int(123)
int(5)

Otherwise, this is correct!

Thanks for catching that. I changed the byte-to-UTF-16 conversion to reject offsets that split a UTF-8 sequence while leaving the referenced offset unchanged. I also added regression coverage for both parse() and parseCurrency(); the focused tests and full Intl suite pass.

Could you please set the error state after failure?
That is, add this line in your tests failure path:

var_dump(intl_is_failure($formatter->getErrorCode()));

expected

bool(true)

This fix is a little different in the master branch since we have the UnicodeString refactors. I will take care.
But just in case, I will wait for David's review before merging.