Problem:
`unpack_string()` validates the declared length against `*size`, the
size *before* `mpack_rtoken()` consumed the token header, rather than
`size2`, the remainder after it. The header is one to five bytes, so any
declared length in the window `(size2, *size]` slips through. The
returned `String` then covers up to five bytes past the end of the
buffer, and `size2 - tok.length` underflows, leaving `*size` near
`SIZE_MAX` so every later unpack call on that entry believes it has an
unbounded buffer.

Reachable from ShaDa, where a history entry ending in a five-byte string
header followed by four bytes is enough, so a corrupted or hostile
`main.shada` triggers it at startup.

Solution:
Check the remainder left after the header.

AI-assisted

Problem:
The magic globals `it`, `describe`, etc., are more trouble than they are
worth.

- Hooking into `after_each` requires `getfenv()` hacks.
- They confuse luals/emmylua, because the top-level `.luarc.json` isn't
  merged with `test/.luarc.json` (apparently a luals limitation?)
- They totally defeat discoverability because the user just has to
  "know" about the various magic symbols.

So they harm DX, which means they serve no purpose at all.

Solution:
- Expose the test API from `testutil`, so tests can call `t.it()`,
  `t.describe()`, etc., in the conventional way.
- Drop `getfenv()` hacks.
- Drop the `setfenv()` injection in `load_chunk`.
- Drop `test/_meta.lua`.

"Error" in error messages is redundant. Just provide the context, don't
say "Error ...".

In the api_info() output:

    :new|put =map(filter(api_info().functions, '!has_key(v:val,''deprecated_since'')'), 'v:val')
    ...

    {'return_type': 'ArrayOf(Integer, 2)', 'name': 'nvim_win_get_position', 'method': v:true, 'parameters': [['Window', 'window']], 'since': 1}

The `ArrayOf(Integer, 2)` return type didn't break clients when we added
it, which is evidence that clients don't use the `return_type` field,
thus renaming Dictionary => Dict in api_info() is not (in practice)
a breaking change.

This also makes shada reading slightly faster due to avoiding
some copying and allocation.

Use keysets to drive decoding of msgpack maps for shada entries.

`FileDescriptor` is already a wrapper around an fd and a buffer.
By allowing to just use the buffer without an fd, it can
already handle in-memory reads.

Specifically, functions that are run in the context of the test runner
are put in module `test/testutil.lua` while the functions that are run
in the context of the test session are put in
`test/functional/testnvim.lua`.

Closes https://github.com/neovim/neovim/issues/27004.

Work on https://github.com/neovim/neovim/issues/27004.