Fix 0-sized getrandom and thread_rng()

[RalfJung]

No description provided.

[RalfJung]

Dang, something is wrong on Windows:

=> captured stdout <=
running 1 test
test test::rng ... ok
test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
running 3 tests
test entropy_rng ... 
=> captured stderr <=
thread 'rustc' panicked at 'assertion failed: `(left == right)`
  left: `8`,
 right: `4`', src\librustc\mir\interpret\value.rs:288:17
stack backtrace:
   0: std::sys_common::alloc::realloc_fallback
   1: std::panicking::take_hook
   2: std::panicking::take_hook
   3: rustc::ty::structural_impls::<impl rustc::ty::context::Lift for rustc::ty::adjustment::AutoBorrow>::lift_to_tcx
   4: std::panicking::rust_panic_with_hook
   5: std::panicking::begin_panic_fmt
   6: std::panicking::begin_panic_fmt
   7: rustc::mir::interpret::value::Scalar<Tag>::to_u64
   8: rustc::mir::interpret::value::Scalar<Tag>::to_usize
   9: rustc_mir::interpret::validity::<impl rustc_mir::interpret::eval_context::InterpretCx<M>>::validate_operand
  10: <rustc_mir::interpret::eval_context::LocalValue<Tag,Id> as core::fmt::Debug>::fmt
  11: rustc_mir::interpret::step::<impl rustc_mir::interpret::eval_context::InterpretCx<M>>::run
  12: miri::eval_main
  13: env_logger::init_from_env
  14: rustc_interface::interface::Compiler::output_file
  15: <miri::MiriCompilerCalls as rustc_driver::Callbacks>::after_analysis
  16: <env_logger::fmt::WriteStyle as core::default::Default>::default
  17: rustc_driver::set_sigpipe_handler
  18: <env_logger::filter::inner::Filter as core::fmt::Display>::fmt
  19: <rustc_driver::pretty::UserIdentifiedItem as core::fmt::Debug>::fmt
  20: rustc_driver::set_sigpipe_handler
  21: _rust_maybe_catch_panic
  22: rustc_driver::set_sigpipe_handler
  23: <std::error::<impl core::convert::From<alloc::string::String> for alloc::boxed::Box<dyn +std::error::Error+core::marker::Sync+core::marker::Send>>::from::StringError as core::fmt::Display>::fmt
  24: std::sys::windows::thread::Thread::new
  25: BaseThreadInitThunk
  26: RtlUserThreadStart
query stack during panic:
end of query stack

[RalfJung]

Looks like the len argument is actually always 32bit on Windows... but how did the tests pass previously then? Almost as if this wasn't even exercised on Windows?

Cc @Aaron1011

[RalfJung]

Oh, I get it... the test actually ICE'd but because of #685 we did not notice.