Work on https://github.com/neovim/neovim/issues/459

Make the copy_object() family accept an optional arena. More than
half of the callsites should be refactored to use an arena later
anyway.

This is both simpler in client code and more effective (always reuse
block hottest in cache)

This reduces the memory overhead for large redraw batches, as a much smaller
prefix of the api object buffer is used and needs to be hot in cache.

In the rare case that exit_event is called from process_close_handles,
it stalls waiting for the process to exit (the routine is currently
underway to do just that). This causes `job_spec.lua` to sometimes
stall.

REJECTED IDEAS:
==============================================================
1. Currently `exit_event` is placed on `main_loop.fast_events`. Would the problem
   be solved by using `main_loop.events` instead?
    - A: Maybe, but it will cause other problems, such as queuing exit_event()
      during "Press Enter..." prompt which may result in the event not being
      processed, leading to another stall.
2. Can we avoid the timer?
    - A: Using a timer is just the easiest way to queue a delayed event without
      causing an infinite loop in the queue currently being processed.
3. Can we avoid the new `exit_need_delay` global...
    1. by using `process_is_tearing_down` instead?
        - A: Can't use `process_is_tearing_down` because its semantics are different.
    2. by checking a similar condition as `process_teardown`? https://github.com/neovim/neovim/blob/f50135a32e11c535e1dc3a8e9460c5b4e640ee86/src/nvim/event/process.c#L141-L142
       ```
       if (!process_is_tearing_down || (kl_empty(main_loop.children) && multiqueue_empty(main_loop.events))) {
         uv_timer_start(&main_loop.exit_delay_timer, exit_delay_cb, 0, 0);
         return;
       }
       ```
        - A: Tried but it did not work (other stalls occurred). Maybe
          exit_event() is called from a source other than
          process_close_handles() and is delayed, the delayed exit_event() will
          be executed before main_loop.events is processed, resulting in an
          infinite loop.

Problem:
- Since c57f6b28d71d #8519, sockets are created in ~/.local/… but XDG
  spec says: "XDG_RUNTIME_DIR: Must be on the local filesystem", which
  implies that XDG_STATE_DIR is potentially non-local.
- Not easy to inspect Nvim-created temp files (for debugging etc).

Solution:
- Store sockets in stdpath('run') ($XDG_RUNTIME_DIR).
- Establish "/tmp/nvim.user/" as the tempdir root shared by all Nvims.
- Make ok() actually useful.
- Introduce assert_nolog().

closes #3517
closes #17093

Note for external UIs: Nvim can now emit multiple "redraw" event batches
before a final "flush" event is received. To retain existing behavior,
clients should make sure to update visible state at an explicit "flush"
event, not just the end of a "redraw" batch of event.

* Get rid of copy_object() blizzard in the auto-generated ui_event layer
* Special case "grid_line" by encoding screen state directly to
  msgpack events with no intermediate API events.
* Get rid of the arcane notion of referring to the screen as the "shell"
* Array and Dictionary are kvec_t:s, so define them as such.
* Allow kvec_t:s, such as Arrays and Dictionaries, to be allocated with
  a predetermined size within an arena.
* Eliminate redundant capacity checking when filling such kvec_t:s
  with values.