Do not require the linker script on std

[jakobjung10]

The script serves no purpose on std but still had to be passed to the linker, excluding linkers without linker script support. probe-rs does not read these binaries, and the setup default now comes from linkme.

Old embedded-test releases must not resolve to the new macros, hence 0.9.0.

[t-moe]

While its true we dont need the linkerfile for the testcase list, Afaik the linker file was also needed to make sure main is not optimized away. (Main comes from a lib instead of from a binary as normally... )

Is this no longer a problem?

[jakobjung10]

On std main is kept by the C runtime startup code, which references it as an undefined
symbol, the same thing that keeps an ordinary fn main. That holds with or without the
linker script.

Checked on x86_64-unknown-linux-gnu with this branch: rustc passes -Wl,--gc-sections and no -T,
nm shows T main in debug and release, and the binary lists and runs its tests. It also
builds and runs on x86_64-pc-windows-msvc, where link.exe can't take a linker script at all.

The script itself never referenced main. EXTERN(__embedded_test_start) and the PROVIDE
alias next to it were added in 27568dd to produce a link error when the script is missing from
the rustflags, and __embedded_test_start is called from main. KEEP(*(.embedded_test.*))
keeps the version symbol and the testcase metadata; on std the test list comes from linkme, so
that isn't needed either. Nothing changes for no_std, which still gets the script.

This also fixes macOS. .embedded_test.meta is an ELF section name and LLVM rejects it on
Mach-O ("mach-o section specifier requires a segment and section separated by a comma"), so on
master --features std doesn't compile for aarch64-apple-darwin at all. Gating that static out
is what makes it build.