Drop the blanket Scala 2 caveat; report the real gap instead

[rochala]

Problem #4 from the printer review, backed by a measurement rather than an impression.

The problem

printSymbolSignatureSafe appended // [Scala 2 — limited type information] to every Scala 2 signature, keyed on source language alone — never on evidence that anything was lost. It fires ~70 times in one cats.data.NonEmptyList lookup, including on def size: Int.

The measurement

Compared scala-library:2.13.16 (pickles) against scala2-library-tasty-experimental_3:3.7.4 — the same standard library in two encodings, so the pairing is exact.

2878 signature lines across 76 symbols. 95.4% byte-identical after normalisation. Zero cases where the Scala 2 side lost a type the Scala 3 side had.

Of the 132 differing lines:

Category	Lines	Worse side
Higher-kinded type args (SeqFactory[List] vs SeqFactory[[A] =>> List[A]])	51	Scala 3
Synthetic $default$N accessors leaking	33	Scala 3
Empty param list (toString() vs toString)	19	Scala 3 (2.13 really declares ())
Implicit-class conversion defs dropped	22	Scala 2
Cosmetics	7	mixed

So Scala 2 renders better roughly 70 lines to 22 — and the one category where it loses is PublicApiFilter.isSyntheticSym dropping symbols, a filter bug a per-signature caveat does nothing about. The TASTy side also failed to load 5 of 81 symbols (None, Predef, AnyVal, Tuple2, SeqOps); the pickle side failed none.

What is actually missing

Documentation. Pickles carry no Scaladoc: 0 of 76 Scala 2 lookups had prose docs versus 70 of 76 from TASTy. Verified directly — cats.effect.IO prints its docstring, cats.data.NonEmptyList prints none.

The change

• Delete the per-signature caveat.
• Reword the once-per-response footer (GetHandler.warnScala2) to Note: Scala 2 artifact — documentation comments are not available. — true for 76/76, unlike the type claim.
• Fix the README support table, which repeated the same inaccurate claim.

No per-signature predicate replaces it. If one is ever wanted, the defensible condition is language-independent — "the printer hit its fallback" — not "the source was Scala 2".

Verification

lib.test 372/372. The two IntegrationTest assertions on the stderr note still hold (the footer still mentions Scala 2, with accurate wording). The TypePrinterTest case that asserted the old caveat is updated to assert its absence plus the real signature.

Follow-ups this surfaced (not in this PR)

Verified against published artifacts, one PR each:

1. Eta-expanded type lambdas — NonEmptyCollection[A, List, [A] =>> NonEmptyList[A]] where Scala 2 prints NonEmptyList; inconsistent within a single line.
2. scala.jdk.CollectionConverters shows 20 class *HasAs* wrappers but 0 of the defs that attach them.
3. scala.Char renders val MaxValue: with a raw U+FFFF and val MinValue: with a raw NUL into Markdown.

🤖 Generated with Claude Code