Binary.md: inlineexport uses exportname but implementations encode exportname'

[LFsWang]

In Binary.md's Instance Definitions, the inlineexport production uses exportname (without prime):

inlineexport ::= n:<exportname> si:<sortidx>    => (export n si)

However, wasm-tools encodes inline export names using exportname' (with the discriminator byte prefix), which is the same encoding used by top-level export and exportdecl:

export     ::= en:<exportname'> si:<sortidx> ed?:<externdesc>?    => (export en si ed?)
exportdecl ::= en:<exportname'> ed:<externdesc>                   => (export en ed)

Example

(component
  (import "f" (func))
  (instance $i (export "hello" (func 0)))
)

wasm-tools parse produces the following bytes for the instance section (section ID 0x05):

 0x17 | 05 0c       | component instance section
 0x19 | 01          | 1 count
 0x1a | 01 01 >00< 05 | [instance 0] FromExports([ComponentExport { name: ComponentExportName("hello"), kind: Func, index: 0, ty: None }])
      | 68 65 6c 6c
      | 6f 01 00   

The 0x00 byte before the name length is the exportname' discriminator. If the spec intended plain exportname (= name = core:name), this byte would not be present and the encoding would instead be:

05 0b          ;; section 5 (instance), 11 bytes
01             ;; vec count: 1 instance
01             ;; instanceexpr flag: 0x01 = inline exports
01             ;; vec count: 1 inline export
               ;; spec no 00 here
05 68 65 6c 6c 6f  ;; name length 5 + "hello" (no discriminator)
01 00          ;; sortidx: sort=0x01 (func), idx=0

[alexcrichton]

I've attempted to address this in #623 by adjusting Binary.md to match what wasm-tools parses today, notably accounting for a historical oddity that we'll want to clean up before 1.0.