wat2wasm --relocatable should record exports in linking symtab

[wingo]

Consider a.c which exports a function that calls something from b.c:

$ cat a.c
void b(void);
__attribute__((export_name("a"))) void a(void) { b(); }

$ cat b.c
void b(void) {}

With clang and wasm-ld, we can make a wasm file:

$ clang -Oz --target=wasm32 -nostdlib -c -o a.o a.c
$ clang -Oz --target=wasm32 -nostdlib -c -o b.o b.c
$ wasm-ld --no-entry -o clang.wasm a.clang.o b.o

Let's say we want to replace a.c with a wat file.

$ wasm2wat -o a.wat a.o
$ wat2wasm --relocatable -o a.wat.o a.wat
$ wasm-ld --no-entry -o wat.wasm a.wat.o b.o

OK great. However:

$ wasm2wat wat.wasm
(module
  (import "env" "memory" (memory (;0;) 2))
  (table (;0;) 1 1 funcref)
  (global (;0;) (mut i32) (i32.const 66560)))

That is to say: dude, where's my export (of a)?

The problem is that in the a.o as produced by clang, the linking and reloc sections look like this:

Custom:
 - name: "linking"
  - symbol table [count=2]
   - 0: F <a> func=1 exported no_strip binding=global vis=hidden
   - 1: F <env.b> func=0 undefined binding=global vis=default
Custom:
 - name: "reloc.CODE"
  - relocations for section: 4 (Code) [1]
   - R_WASM_FUNCTION_INDEX_LEB offset=0x000004(file=0x000077) symbol=1 <env.b>
Custom:
 - name: "producers"

Code Disassembly:

000075 func[1] <a>:
 000076: 10 80 80 80 80 00          | call 0 <env.b>
 00007c: 0b                         | end

Whereas as produced by wat2wasm, they look like this:

Custom:
 - name: "linking"
  - symbol table [count=1]
   - 0: F <env.b> func=0 undefined binding=global vis=default
Custom:
 - name: "reloc.Code"
  - relocations for section: 4 (Code) [1]
   - R_WASM_FUNCTION_INDEX_LEB offset=0x000004(file=0x000063) symbol=0 <env.b>

Code Disassembly:

000061 func[1] <a>:
 000062: 10 80 80 80 80 00          | call 0 <env.b>
 000068: 0b                         | end

I know that on #1137 there is a question about whether to keep up the functionality of --relocatable at all, but I think that with a couple of bug-fixes it can be usable enough for me, so I will take a look at fixing this unless there's an objection.

[wingo]

To fix this, I could either add an "exported" flag to the various IR nodes, and use it when recording symbol references in the binary writer, or I could precompute a side table in the binary writer if we are doing --relocatable. I guess it's easier to reason about a side table, compared to a mutable flag, so I'll take that approach.