To avoid confusion, on PHP <7.4 the output is:
array(15) {
  [0]=>
  string(0) ""
  [1]=>
  string(0) ""
  [2]=>
  string(0) ""
  [3]=>
  string(0) ""
  [4]=>
  string(0) ""
  [5]=>
  string(0) ""
  [6]=>
  string(0) ""
  [7]=>
  string(0) ""
  [8]=>
  string(0) ""
  [9]=>
  string(0) ""
  [10]=>
  string(0) ""
  [11]=>
  string(0) ""
  [12]=>
  string(0) ""
  [13]=>
  string(0) ""
  [14]=>
  string(0) ""
}

Output on > 7.4

array(15) {
  [0]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [1]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [2]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [3]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [4]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [5]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [6]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [7]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [8]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [9]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [10]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [11]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [12]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [13]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
  [14]=>
  array(2) {
    [0]=>
    string(0) ""
    [1]=>
    int(-1)
  }
}

Actually the output in 7.4 seems to be correct as before the offset was completely missing. This was fixed in Bug #73948

also - the output in 7.4 actually matches the documentation better - don't you agree?

What's more, this is mentioned by NEWS and UPGRADING so it's not like this change has been forgotten...

This is something that I would like to see highlighted because it can affect all templating engines who has some Lexers e.g.: Twig

It will be documented in the function's Changelog and the 7.4 migration guide. That is how these sorts of changes are always handled. Anyone blindly upgrading their codebase to 7.4, or any new version series, is shooting themselves in the foot.

Is that acceptable? If not then what are you asking for to happen?

This specific case doesn't seem to be mentioned in UPGRADING (it currently only discusses the UNMATCHED_AS_NULL case), so that at least should be added.

As Nikita mentioned only UNMATCHED_AS_NULL is covered and OFFSET_CAPTURE was missed (probably other cases are also missed)