Fix context impl cleanup.

[hidmic]

Precisely what the title says. Otherwise, SIGSEGV ensues if a subsequent (re)initialization fails halfway through.

[ivanpauno]

Otherwise, SIGSEGV ensues if a subsequent (re)initialization fails halfway through.

Do we provide that guarantee?

Just to double check, this is what I understand from your comment:

rmw_context_t context = rmw_create_context(...);
rmw_destroy_context(context); // this one fails
rmw_destroy_context(context); // retrying after failing segfaults

[hidmic]

Just to double check, this is what I understand from your comment:

I meant subsequent context impl (perhaps that's the bit that's missing) (re)initialization failure after a preceding (often implicit) cleanup may result in SIGSEGV. That can happen if, within rmw_create_node, rmw_context_impl_t::init() fails halfway through twice in a row.

[ivanpauno]

I meant subsequent context impl (perhaps that's the bit that's missing) (re)initialization failure after a preceding (often implicit) cleanup may result in SIGSEGV. That can happen if, within rmw_create_node, rmw_context_impl_t::init() fails halfway through twice in a row.

I see, it's super weird though.

[hidmic]

CI up to rcl:

• Linux
• Linux-aarch64
• macOS
• Windows (warnings fixed by Change rcutils_fault_injection_set_count to use int64_t rcutils#283)