feat: ported to Vita target

[nikarh]

Fixes #160

Notes

• signal-hook does not compile and won't work on Vita, pthread_kill is a stub in Vita newlib
• Pipes in Vita newlib do not guarantee that after the write completes, the payload is immediately available on the other side of the pipe. This PR temporarily removes the NONBLOCK flag from the read side of the pipe when wait detects notified before polling

[notgull]

Can we add a test for this in CI? Just a cargo check would suffice.

Also, do you have a way of testing this on the actual Vita software?

[nikarh]

Can we add a test for this in CI? Just a cargo check would suffice.

Also, do you have a way of testing this on the actual Vita software?

Running tests requires an actual piece of hardware (no way to run in qemu, since Vita runs a proprietary OS that demands specific hardware). Cargo check though is absolutely doable, will add it.

Currently, I run tests on a physical device using cargo-vita tool which does these things in a single command:

• Compiles rust test code and links it with VITASDK toolchain gcc to an ELF (this is essentially cargo build with a bunch of flags against the target, linker is defined in the target spec)
• Transforms the usual ELF into Sony ELF (by calling a number of tools from VITASDK toolchain)
• Uploads the Sony ELF to Vita via FTP (usually all devs who target Vita run a userspace module on the device that keeps an FTP server and a TCP command server)
• Sends a "run this app" command to the TCP command server

At the same time, I run another plain text TCP socket server on my computer, and Vita runs a module that pipes stdin and stdout to that TCP socket, making it possible to see test results.

[nikarh]

Fixed the comment, added check to CI and enabled CI check for tier 3 nightly targets

[notgull]

I'm assuming the tests pass when you run it on actual hardware, right?

[nikarh]

I'm assuming the tests pass when you run it on actual hardware, right?

Yes, I run them on the actual hardware, and all of them pass (except for the many_connections due to fd limits on the system, but it also passes if I reduce 100 to 60)

[notgull]

Thanks! NetBSD failure looks spurious

[taiki-e]

Instead of adding +nightly to all cargo command calls, I recommended using && rustup default ${{ matrix.rust }} here.