fix bare optional/star/plus quantifiers on rule references

[datduyng]

Additive only: bare ?/*/+ now parse like grouped form; required refs, grouped (), and existing patterns unchanged.

We currently have many cases of "?" usage which represent optionality but are treated as literals.

<VideoPhrase> = <Polite>? <CreateVerb> <VideoTarget> <VideoContent>? | <VideoTarget> <FromFiles> | <NeedPhrase> <VideoTarget>;

TypeAgent/ts/packages/agents/video/src/videoSchema.agr

Line 27 in 2596224

<VideoPhrase> = <Polite>? <CreateVerb> <VideoTarget> <VideoContent>? | <VideoTarget> <FromFiles> | <NeedPhrase> <VideoTarget>;

https://github.com/search?q=repo%3Amicrosoft%2FTypeAgent+%22%3E%3F%22+path%3A**%2F*.agr&type=code

On main right now, we have below behavior. The PR fix this so that "?" is not treated as a literal

BARE grammar:

<Polite> = please | could you | can you | kindly;
<Start> = <Polite>? open outlook → "matched";

BARE Start expressions:

{
  "type": "ruleReference",
  "name": "Polite",
  "optional": false,
  "repeat": false
},
{
  "type": "string",
  "value": [
    "?",
    "open",
    "outlook"
  ]
}

[datduyng]

Before

After

[datduyng]

Will work offline and review the design with @curtisman offline first - putting this in draft for now

[datduyng]

Parked for design debate.

Immediate unblock (grammar-only, no language change): #2775

• rewrites bare `?` → `( )?` so optionals match under current semantics
• RFC one-pager for Design A (group-gated, status quo) vs Design B (reserved `?/*/+`, this PR's direction): `ts/packages/actionGrammar/docs/rfc-agr-quantifier-semantics.md`

Leaving this PR draft until Curtis picks A vs B.