Fix tests for bidirectional isolation

[eemeli]

The current tests do not account for the bidirectional isolation of the formatted results. This adds an option bidiIsolation to the tests, with values 'default' and 'none' corresponding to the Default BiDi Strategy and to a strategy of not applying any isolation.

Most of the existing tests follow the 'none' strategy, and have been left unchanged except for noting this state of affairs. The bidi.json tests have been updated to use the 'default' strategy, as that seemed appropriate.

A new type: 'bidiIsolation' part is added, and included in the u-options.json expParts results as appropriate; some bugs in those tests have also been addressed.

[eemeli]

We may need to drop this test, or more explicitly define whether :string applies introspection to its value to determine its directionality. As we don't forbid introspection, an implementation could use either an RLI or FSI here.

Presuming that the introspection would follow UAX#9, the user's perception of the results would be the same, so my preference would be to make the character-level output predictable by requiring that no introspection is done in the absence of a u:dir override or an operand that includes an explicit direction.

[aphillips]

I puzzled with it for a while but think this is right. The prevalence of FSI/PDI wrappers even in a purely LTR context makes me itch.

[catamorphism]

I guess in this case, the direction of the message is determined to be unknown? Which is a little surprising to me, since the only bidi control character appears in one of the variable names, which I would expect not to have an effect on the formatted result of the message. Maybe what's throwing me off is that the explanation of the "Default Bidi Strategy" at the end of this section doesn't say how the message directionality is determined?

[eemeli]

The bidi controls in the syntax source don't matter here at all, and the message has LTR direction because of the en-US locale. It's the placeholder's direction that is unknown. For it to be known, we'd need to introspect the 1 string with something like UAX#9 P2, which will ultimately be done by the rendering engine, and even if we do, the direction would still be unknown because European numbers are only weakly directional. And so this is appropriately reflected by FSI/PDI isolating it.

This is much the same thing as I mention in #917 (comment), but for LTR instead of RTL content.

Maybe what's throwing me off is that the explanation of the "Default Bidi Strategy" at the end of this section doesn't say how the message directionality is determined?

Good point. We may want to add a mention under Formatting Context about deducing the message direction from the locale if it's not explicitly given.

[aphillips]

If not explicitly provided, the direction of the message should be computed from the locale. As @eemeli notes, controls in the source don't matter. As noted elsewhere, you can't just look at the rendered placeholder to determine its desired direction--even strongly directional characters do not necessarily provide an accurate signal.

[catamorphism]

https://github.com/unicode-org/conformance/blob/main/schema/message_fmt2/testgen_schema.json will need to be updated to reflect this change.