DefaultValue will return undefined for externref as before, but
for other nullable reference types will return null.

For non-nullable reference types, operations like grow and set
should error if the value to set is missing.

Also adds WPT tests for the error cases. This requires some
updates to wasm-module-builder.js as well.

Fixes issue WebAssembly#501

Instead of allowing ToWebAssemblyValue to throw a TypeError, explicitly catch
any exceptions and throw a WebAssembly.LinkError.

Instead of allowing ToWebAssemblyValue to throw a TypeError, explicitly catch
any exceptions and throw a WebAssembly.LinkError.

Co-authored-by: Andreas Rossberg <rossberg@mpi-sws.org>

[js-api][test] Refine DefaultValue for new reference types

[js-api] Throw consistent exns in "read the imports"

Test that validation allows `call_indirect` to have a type immediate that is a
supertype of or unrelated to the table element type and that the runtime
behavior is correct in these cases.

The first operand for `ref.cast` should be a reference type.
However, the code example in `MVP.md` seems to have a heaptype
which is invalid.
(Assuming the `$C` is defined as follows based on the context)

```
(module
  (type $A (sub (struct)))
  (type $B (sub $A (struct (field i32))))
  (type $C (sub $B (struct (field i32 i32))))
  (func $castBtoC (param $x (ref $B)) (result (ref $C))
    (ref.cast $C (local.get $x))
  )
)
```

```
$ wasm -d test.wat
test.wat:6.15-6.17: syntax error: unexpected token
```

(Where `wasm` is a reference interpreter built from
WebAssembly/gc@0ba84fb)

Fix an example code for ref.cast in MVP.md

[test] Test `call_indirect` subtyping more thoroughly

It produces a non-null `(ref i31)` not an `i31ref` aka `(ref null i31)`.

Fix typing of `ref.i31` in the appendix

I noticed that Wasmtime was passing the spec tests despite having bits that were
known to be unimplemented. This should help the tests exercise those corners of
the spec.

Notably:

* Setting `i31ref` globals.

* Initializing tables and globals with `(ref.i31 (global.get $g))`.

* Table operations on `i31ref` tables.

* Accessing `anyref` globals and tables that are actually `i31ref`s. This is
  interesting to exercise outside of general subtyping because we have different
  paths in our inline GC barriers for `anyref`s that are actually `i31ref`s.

Add additional `i31ref` spec tests

The prose incorrectly used "larger than or equal to" when it should have used
just "larger than." The formalism was already correct.

Fixes WebAssembly#536.

…py-bounds

[Spec] Fix bounds check prose for array.fill and array.copy

Fix wrong type expansion

Co-authored-by: Manos Koukoutos <manoskouk@google.com>

More fixes and tests around element segments

Copy-paste bug. Don't repeatedly assert the first element's value, but each element in the table.