Add more benchmarks:

[dmitry-timofeev]

Overview

• Push/PopLocalFrame overhead benchmark — such operations
  are required if one attaches a long-running native thread
  to the JVM because there is no 'return-from-native-method'
  event when created local references are freed, hence no way for
  the JVM to know that the local references are no longer used
  in the native code.

• A benchmark of the overhead of attaching and detaching
  a native thread. It is expectedly HUGE — two orders of magnitude
  higher than calling a single Java method using unchecked APIs
  (e.g., jni_call_static_unsafe).

Results

$ cargo +nightly-2018-08-01 bench --features=invocation,backtrace 
   …

running 16 tests
test tests::jni_call_object_method_safe          ... bench:       2,049 ns/iter (+/- 64)
test tests::jni_call_object_method_unsafe        ... bench:         276 ns/iter (+/- 25)

test tests::jni_call_static_method_safe          ... bench:       2,577 ns/iter (+/- 65)
test tests::jni_call_static_method_unsafe_jclass ... bench:         293 ns/iter (+/- 11)
test tests::jni_call_static_method_unsafe_str    ... bench:         840 ns/iter (+/- 32)

test tests::jni_get_java_vm                      ... bench:          11 ns/iter (+/- 0)

test tests::jni_new_global_ref                   ... bench:         195 ns/iter (+/- 1)

test tests::jni_new_object_by_id_jclass          ... bench:         251 ns/iter (+/- 13)
test tests::jni_new_object_by_id_str             ... bench:         765 ns/iter (+/- 65)
test tests::jni_new_object_jclass                ... bench:       1,917 ns/iter (+/- 47)
test tests::jni_new_object_str                   ... bench:       2,561 ns/iter (+/- 84)

test tests::jni_noop_with_local_frame            ... bench:          96 ns/iter (+/- 0)
test tests::jvm_noop_attach_detach_native_thread ... bench:      26,616 ns/iter (+/- 2,445)

test tests::native_arc                           ... bench:          11 ns/iter (+/- 0)
test tests::native_call_function                 ... bench:           2 ns/iter (+/- 0)
test tests::native_rc                            ... bench:           3 ns/iter (+/- 0)

test result: ok. 0 passed; 0 failed; 0 ignored; 16 measured; 0 filtered out

Definition of Done

• There are no TODOs left in the code
• Change is covered by automated tests
• The coding guidelines are followed
• Public API has documentation
• This change is not breaking or mentioned in the Changelog
• The continuous integration build passes