GOOS=js GOARCH=wasm has not compiled for 3 days, and the test that catches it is behind an env var no workflow sets #682

GOOS=js GOARCH=wasm does not compile, and has not for about three days:

$ CGO_ENABLED=0 GOOS=js GOARCH=wasm go build ./internal/vm
internal/vm/errno.go:101:35: undefined: syscall.ENOTRECOVERABLE
internal/vm/errno.go:108:35: undefined: syscall.EOWNERDEAD
internal/vm/errno.go:120:35: undefined: syscall.ETXTBSY

errno.go carries no build tags, so it compiles for every target. wasip1
defines those three names; js does not. This breaks cmd/rbgo, the cmd/wasm
playground and rbgo build --closed --target wasm — the whole browser story.

It is a regression, and the bisect is done: internal/vm and cmd/rbgo
both build for js/wasm at 68cb53a, the commit before d498ddc (the Errno
subsystem, #629, 2026-09-23). A 95 MB module came out.

Two screens, both of which failed open

This is the more useful half of the report.

1. ci.yml gates wasip1 only. There is no js/wasm lane, so no PR could
   have caught it.

2. TestClosedWasmBuildIntegration does catch it — and cannot run. It sits
   behind RBGO_BUILD_IT=1, and:

   $ grep -rn 'RBGO_BUILD_IT' .github/workflows/
   (no output)
   

   Nothing sets it, anywhere. Run with it set, the test fails with exactly the
   three errors above. The guard was correct, complete, and unreachable.

So the fix is two parts, and the second matters more than the first:

• build-tag errno.go (or guard the three constants per-GOOS);
• give js/wasm a CI lane, and either wire RBGO_BUILD_IT=1 into a lane or stop
  pretending that test is a gate. A guard behind an env var that no workflow sets
  is documentation, not a check.

A third, smaller finding from the same pass: rbgo build only works from a
checkout — run elsewhere it fails with stat .../cmd/rbgo: directory not found. The docs never said so.