Add stdlib test importing every stdlib module + fix stdlib

[jiribenes]

Resolves #614 by adding a stdlib test that imports every single (common) stdlib module.

Let's add ACME (in our context "All Common Modules in Effekt" :)) to the stdlib tests: just import and compile all modules.

This revealed 3 different bugs in stdlib + codegen bugs on LLVM (moved to #801).

[jiribenes]

Ideal follow-up: make CI so that on change in /libraries/common/, we check that the acme.effekt file contains every module in that folder, recursively (just generate it again and diff). If not, fail and spam the PR.
EDIT: Done. ACME CI now exists.

[jiribenes]

Note

Moved to #801

New LLVM bugs, cc @phischu:

$ EFFEKT_DEBUG=1 effekt --backend=llvm --ir-write-all --debug --optimize examples/stdlib/json.effekt
opt: ./out/json.ll:4001:9: error: multiple definition of local value named 'd_24_132_3_22162'
        %d_24_132_3_22162 = insertvalue %Neg %vtable_temporary_865, %Object %closure_861, 1
        ^

$ EFFEKT_DEBUG=1 effekt --backend=llvm --ir-write-all --debug --optimize examples/stdlib/buffer.effekt
Instruction does not dominate all uses!
  %buffer_24_4387 = insertvalue %Neg %vtable_temporary_1257, ptr %closure_1239, 1
  call void @shareNegative(%Neg %buffer_24_4387)
Instruction does not dominate all uses!
  %buffer_24_4387 = insertvalue %Neg %vtable_temporary_1257, ptr %closure_1239, 1
  store %Neg %buffer_24_4387, ptr %buffer_24_4387_pointer_1255, align 8, !noalias !0
Instruction does not dominate all uses!
  %buffer_24_4387 = insertvalue %Neg %vtable_temporary_1257, ptr %closure_1239, 1
  call void @eraseNegative(%Neg %buffer_24_4387)
opt: ./out/buffer.ll: error: input module is broken!

EDIT: I think it's the same problem, actually.
For negative types, we generate:

define tailcc void @xyz_clause_1(...) { ... }
define tailcc void @xyz_clause_2(...) { ... }

which always share what would be:

; new xyz_clause_6769, 0 parameters
        %xyz_6769_pointer_492 = getelementptr {%reference, %reference, %neg, %pos, i64}, %environment %environment_489, i64 0, i32 2
        %xyz_6769 = load %neg, ptr % xyz_6769_pointer_492, !noalias !2

so the same value is defined under the same name in multiple different functions.

[jiribenes]

I think this is semantically correct, we use it as a circular buffer, so we don't really need to invalidate the previous value -- we'd have overwritten it.

[jiribenes]

I'll merge this since it's a change that I really want in the next release, any nitpicks can be fixed in a separate PR