Compile BuiltinData transparently in the plugin

[Unisay]

Alternative to #7719 for fixing #7716 — the two PRs are mutually exclusive; whichever lands, the other closes.

The plugin treats BuiltinData as representationally transparent: PlutusCore.Data.Data compiles to the builtin data type (same as BuiltinData), the BuiltinData constructor compiles to the identity, and a single-alternative case on it binds both the case binder and the field to the scrutinee. GHC unwrapping the wrapper — the worker/wrapper unboxing that exposed Data in join point types and crashed the plugin with Unsupported feature: Type constructor: GHC.Prim.Addr# — then becomes harmless instead of fatal. Details in Note [Transparent BuiltinData] in PlutusTx.Compiler.Expr.

The problematic Core shape, from the unfolding of the useTwiceData regression test (condensed) — the source mentions only BuiltinData, yet the join point's type signature exposes Data:

useTwiceData = \ (bd :: BuiltinData) ->
  case bd of bd1 { BuiltinData ipv ->
    let $j :: PlutusCore.Data.Data -> BuiltinUnit   <join 1>
        $j = \ (ipv2 :: PlutusCore.Data.Data) -> ...
    in ... }

Unlike the second-constructor approach, BuiltinData itself is untouched and costs don't move: no budget golden changes under either GHC version. Large-script UPLC goldens (cardano-constitution, Ed25519) shift shape through different CSE choices at identical budget and CBOR size.

Stage violations that previously died at the Cannot construct a value of type: BuiltinData check now surface as a missing-unfolding error; the last commits port #7719's improved diagnostics to that path (see the StageViolation/builtinData goldens).