fix(strings): count and slice strings by Unicode code point

str_length and str_slice used UTF-8 byte offsets, so any input with
multi-byte characters produced wrong results: str_length("cafe") with
an accented final character returned 5 instead of 4, and str_slice(0,
4, ...) on the same string returned an empty string because the byte
index fell inside the encoding of the last character. This also broke
the higher-level helpers built on top of them (str_find, str_contains,
str_replace).

Count characters with chars().count() and index str_slice by character.
Out-of-range slices still return an empty string, so behavior for ASCII
input is unchanged.

examples/tests/strings.nbt

@@ -1,11 +1,19 @@
 assert_eq(str_length(""), 0)
 assert_eq(str_length("foo"), 3)
 
+# Length and slicing operate on Unicode code points, not on UTF-8 bytes
+assert_eq(str_length("café"), 4)
+assert_eq(str_length("❤"), 1)
+
 assert_eq(str_slice(0, 5, "hello world"), "hello")
 assert_eq(str_slice(6, 11, "hello world"), "world")
 assert_eq(str_slice(0, 0, "hello world"), "")
 assert_eq(str_slice(0, 100, "hello world"), "")
 
+assert_eq(str_slice(0, 4, "café"), "café")
+assert_eq(str_slice(3, 4, "café"), "é")
+assert_eq(str_slice(0, 3, "café"), "caf")
+
 assert_eq(chr(65), "A")
 assert_eq(chr(97), "a")
 assert_eq(chr(0x2764), "❤")