Fixes #15147 and fixes #15497. Also sketch "subdir" caching. Currently
this only caches whether an rtp entry has a "lua/" subdir but we could
consider cache other subdirs potentially or even "lua/mybigplugin/"
possibly.

Note: the async_leftpad test doesn't actually fail on master, at least
not deterministically (even when disabling the fast_breakcheck
throttling). It's still useful as a regression test for further changes
and included as such.

Do not copy a lot of lua strings (dict keys) to just strequal() them
Just compare them directly to a dedicated hash function.

feat(generators): HASHY McHASHFACE

These functions do not involve msgpack. Initially the nvim api was
sometimes called the "msgpack API", but entry points from vim script
and lua are equally valid (and don't need to reference "msgpack")

Note: the reason for removing them is not that there after this refactor
is no use of them, but rather that having them available is an
anti-pattern: they manange an _extra_ heap allocation which has
nothing to do with the functionality of the map itself (khash
manages the real buffers internally). In case there happens to
be a reason to allocate the map structure itself later, this
should be made explicit using xcalloc/xfree calls.

vim.api.nvim_chan_send(vim.api.nvim_open_term(0), io.open("/path/to/smile.cat", "r"):read("*a"))

* improve error message to make it actionable