Perform the following optimisations:

• Don't linearise Latin-1 strings.
• Use InlineCharBuffer to avoid heap allocations for small strings.
• Return the result of Utf16ValidUpTo from IsStringWellFormedUnicode so we can skip the initial characters when calling EnsureUtf16ValiditySpan.
• Don't try to deflate non-wellformed strings, because they will always contain the replacement character and therefore can't be stored as Latin-1 strings.

Perform the following optimisations:

• Don't linearise Latin-1 strings.
• Use InlineCharBuffer to avoid heap allocations for small strings.
• Return the result of Utf16ValidUpTo from IsStringWellFormedUnicode so we
  can skip the initial characters when calling EnsureUtf16ValiditySpan.
• Don't try to deflate non-wellformed strings, because they will always contain
  the replacement character and therefore can't be stored as Latin-1 strings.

Also add some additional tests to cover string types which aren't tested in
the test262 tests.

https://hg.mozilla.org/mozilla-central/rev/777551a9013e