spectest: reproduce segfault

[evacchi]

So this was definitely a journey :D This PR should fix ... for the old compiler, but really it addresses an issue that is unrelated to the file cache, or the compiler for that matter. It was indeed related to the finalizer, but it is much more articulated than you'd think.

Indeed, the problem is a lifecycle issue that involves finalizers, and it definitely relates to invoking code that is no longer mmapped; however, it was not due to a module being unmapped while executing, but it is a module doing a call_indirect on a module that was collected.

The spect tests generally do not fail because the GC is not performed at each iteration, but if we force it to collect at the end of each unit test (ab4f3d8) the spec tests will fail very predictably. In fact, by allowing a bit of output, we learn that one of the tests in linking.wast enters (->) a code segment with a given address (in this case 0x12aba0000) in assert_uninstantiable, then leaves it (<-), but then this segment is immediately released (!!), i.e. finalized. This is because

        === RUN   TestFileCache_compiler/spectest/linking.wast/assert_uninstantiable/line:371
        2024/02/28 16:54:11 -> 12aba0000
        2024/02/28 16:54:11 <- 12aba0000
        2024/02/28 16:54:11 !! 12aba0000
        === RUN   TestFileCache_compiler/spectest/linking.wast/assert_return/line:387
        2024/02/28 16:54:11 -> 12ab90000
        2024/02/28 16:54:11 <- 12ab90000
        === RUN   TestFileCache_compiler/spectest/linking.wast/assert_return/line:388
        2024/02/28 16:54:11 -> 12ab90000
        unexpected fault address 0x12aba0000
        fatal error: fault
        [signal SIGSEGV: segmentation violation code=0x2 addr=0x12aba0000 pc=0x12aba0000]