Patch margin index splits Unicode surrogate pairs unexpectedly

When the function diff_match_patch.prototype.patch_addContext_ adds context to a patch, it increments/decreases the index by a constant, Patch_Margin = 4. However, since JavaScript's substring function operates with UTF-16 code unit indexing, there's a chance that Patch_Margin may split a Unicode surrogate pair.

Consider the following example:

import diff_match_patch from "diff-match-patch";

console.log(
  JSON.stringify(
    new diff_match_patch().patch_make(" **a", " **")[0].diffs[0][1],
  )
);

The output is "\uddee **" ( corresponds to "\ud83e\uddee").

If you attempt to use diff_match_patch.patch_obj.prototype.toString on this patch, it leads to a crash. encodeURI will throw a URIError if URI contains a lone surrogate.

import diff_match_patch from "diff-match-patch";

const diff = new diff_match_patch();

console.log(
  JSON.stringify(
    diff.patch_toText(diff.patch_make(" **a", " **")) // URIError: URI malformed
  )
);

A straightforward solution might involve adding a verification step after applying Patch_Margin to ensure the indices remain valid. I can start a PR, but I've noticed that Patch_Margin is used in many places, and I'm unsure about the best way to make changes.