reels demo terminates on resize

[dankamongmen]

If we resize the notcurses-demo window while the reels demo is running, the demo exits immediately. It ought instead take the resize into account, and redraw.

[dankamongmen]

We're exiting out in the handle_input() when we come back to the top of the loop, and this is happening when we press most buttons :/.

[dankamongmen]

Hrmm. This is somewhat complex, and has ramifications for #373. Demos don't call notcurses_getc() directly, but instead demo_getc() and demo_getc_nblock(). This is necessary, since the demo controller always grabs input, looking for top-level controls.

Note that if we enter raw mode, we are planning to drain input internally to notcurses (to properly handle e.g. Ctrl+C without latency), so we'll see the same situation.

The problem is that we then can't use poll() to determine whether input is available. It's already been removed. The reel demo needs both watch input and an eventfd (written to by way of ncreel_touch() from tablet_thread()). It's POLLIN on stdin thus is borked (it only picks up on a later event). Hrmm.

My initial idea is that such a scheme can use its own eventfd/pipe. Kinda heavyweight, but it works, and also allows us to handle e.g. multibyte input entirely in the primary receiver...

[dankamongmen]

We can solve this for the purpose of reel by going ahead and doing a trial run of the eventfd idea, connected from the demo input handler out to reel.

[dankamongmen]

No longer happening for most keypresses, but still happening for NCKEY_RESIZE. Odd.

[dankamongmen]

Ahhh, the remaining NCKEY_RESIZE was blowing up due to poll() being interrupted by SIGWINCH. I've wrapped EINTR, and things now all work properly.