Bidi tests rely on assumptions about standard functions

[catamorphism]

I noticed that the tests where the default bidi strategy is enabled expect numbers not to be isolated, but expect strings to be isolated. For example:

https://github.com/unicode-org/message-format-wg/blob/main/test/tests/bidi.json#L91

    {
        "description": "function-expression = \"{\" o function *(s attribute) o \"}\"",
        "src": "{1 \u200E :number \u200F}",
        "exp": "1"
    }

and

https://github.com/unicode-org/message-format-wg/blob/main/test/tests/u-options.json#L143

 {
      "locale": "ar",
      "src": "أهلاً {بالعالم :string}",
      "exp": "أهلاً \u2068بالعالم\u2069"
    }

In both cases, the function (:number or :string) is called without a u:dir or u:locale option, so I would expect either both to be isolated, or neither to be isolated. The assumption seems to be that the result of a :number gets its directionality from the underlying locale, while the result of a :string has unknown directionality (in the absence of u:dir). This should be documented in the spec.

[aphillips]

I agree that this might be a bug in the first test quoted, or at least a gap. If msgdir and fmt are both LTR, then you'd expect the output shown. But we don't state the locale or specify what :number will do in this case. Without those, the output should have FSI/PDI around the 1. More documentation is probably needed.

I'll note that the test in question is trying to test whether strong bidi marks can appear in the expression without causing errors.

[aphillips]

Researching more deeply.

The first test should have a locale so that it's direction is LTR. The default isolation strategy does not isolate LTR in LTR automatically. The second test is fine as is. I'll do a PR to fix the first one.

[aphillips]

... and when I looked at the file, I found that the locale is set in the test header, so no change is needed.