Preserve a Team constant written in a Color slot instead of wrapping it in Color(...)

[e54-bot]

Goal

A Team constant written where a Color parameter is expected keeps its authored form through parse and emit, so it is not charged an extra element.

Context

OWBastion/Bastion passes a Team value in a Color position (playEffect(..., DynamicEffect.ECHO_STICKY_BOMB_EXPLOSION, Team.2, eventPlayer, 200)). Pinned OverPy 9.7.10 writes it as written, Team 2, and the client imports that. workshop-rs writes Color(Team 2):

input:  Play Effect(All Players(All Teams), Echo Sticky Bomb Explosion Effect, Team 2, Event Player, 200);
emit:   Play Effect(All Players(All Teams), Echo Sticky Bomb Explosion Effect, Color(Team 2), Event Player, 200);

The project has 14 such calls (Team 1 four times, Team 2 ten times); OverPy also writes Color(Team N) in 14 other Play Effect calls and 28 Create Effect calls where the source used the Color constant. The client counts the Wright build of this project 20 elements above the OverPy build (30090 against 30070), and OverPy's own accounting charges the wrapper one element, so these 14 account for most of the difference. The remaining 6 are not isolated; candidates are the single False/True count difference and the While condition spelling below.

While((a) && (b) && (c) && (d)) in OverPy's output is written as While(And(And(And(a, b), c), d)) by workshop-rs (rule [Hero/D.Va] Sync remech charge); whether the client counts them the same is unknown.

Scope

• Keep the authored spelling of a value whose type differs from the parameter's when the client accepts it (evidence: OverPy's output is imported), instead of resolving it to the parameter's enum member.
• Cover it with a regression test in en-US and zh-CN.
• Record whether the infix boolean chain and the nested And cost the same in the client.

Non-goals

• Element-count model (separate issue).
• Warning about the type slip in the source (a source-language diagnostic).

Acceptance criteria

• Parse then emit of the Play Effect line above reproduces the input.
• On the project above the Wright build's Color( count equals OverPy's.

Dependencies

Related: #302, ADR-0014, ADR-0015.

[indent]

Verified on current `main` (v0.9.1, includes #307):

Acceptance criterion 1 — parse/emit round-trip of the exact example from this issue:

```
input: Play Effect(All Players(All Teams), Echo Sticky Bomb Explosion Effect, Team 2, Event Player, 200);
emit: Play Effect(All Players(All Teams), Echo Sticky Bomb Explosion Effect, Team 2, Event Player, 200);
```

Reproduces the input exactly; no longer wrapped in `Color(...)`.

Acceptance criterion 2 — full round-trip of `tests/fixtures/real-projects/bastion.ow` (the pinned OverPy-generated real-project fixture): `Color(` count is 130 in the source and 130 in the re-emitted output. The specific 6 bare / 1 wrapped `Echo Sticky Bomb Explosion Effect` split called out in this issue also matches exactly, and explicit `Color(Team N)` calls elsewhere in the same project (e.g. `Create Effect(..., Color(Team 1), ...)`) are still preserved as Color identities, confirming validation isn't conflating the two.

Both acceptance criteria are met, so I'm closing this.

The Scope's third bullet — whether `While(a && b && c && d)` and the emitted `While(And(And(And(a, b), c), d))` cost the same in the client — is still unrecorded, same as noted in #307's "Runtime evidence still needed" section. That needs an actual Overwatch client capture of the `[Hero/D.Va] Sync remech charge` rule (or an equivalent isolated case), which isn't available in this environment. The element-count model added in #308 (a separate issue, #305) matches OverPy's total for this project to within a handful of elements, all attributed to documented `End`-spelling differences elsewhere, none attributed to this rule's `And` nesting — suggestive but not a direct measurement. If this is worth tracking on its own, it should be a new issue scoped to "capture client element cost for infix vs nested boolean forms," since it depends on external client access rather than repo work.