tests: Add NVMe-MI recovery test

Exercise the NVMeMi endpoint recovery implementation by exploiting the
link seam for libnvme{,-mi}.

To do this we first reuse the headers provided by libnvme with a
partial-dependency declaration in the test. From there we provide our
own (largely neutered) implementation of the APIs to link against.

With this in-place we can trivially recreate the error discussed in[1]
by returning erroneous values from the libnvme APIs. The higher-level
consequence is we drive the NVMeMi implementation through the endpoint
recovery process while trying to optimize the connection. At that point
we can use the usual GMock APIs to define the behaviors we want from
MctpDevice and MctpEndpoint during the recovery process.

[1]: https://github.com/CodeConstruct/dbus-sensors/issues/14

An interesting future direction is to allow mocks to be injected behind
the libnvme{,-mi} link seam. Adding a method to inject mocks enables
describing arbitrary NVMe scenarios in the usual gmock style, which has
the potential to drastically increase test code coverage for nvmesensor.

Returning to the change at hand, note that the test case causes valgrind
to report a leak, however this appears to be from the gmock error report
itself. Given we're working to resolve the gmock error, the leak will
disappear with it:

```
...
Mock MCTP Device] MCTP connection is not established
error reading ctemp from subsystem, reason:Transport endpoint is not connected
Sensor bar reading error!
../tests/test_nvme_mi_recovery.cpp:63: Failure
Mock function called more times than expected - returning directly.
    Function call: recover()
         Expected: to be called between 1 and 3 times
           Actual: called 5 times - over-saturated and active

[Mock MCTP Device] MCTP connection is not established
error reading ctemp from subsystem, reason:Transport endpoint is not connected
../tests/test_nvme_mi_recovery.cpp:63: Failure
Mock function called more times than expected - returning directly.
    Function call: recover()
         Expected: to be called between 1 and 3 times
           Actual: called 6 times - over-saturated and active

status else
poll loop has been canceled
../tests/test_nvme_mi_recovery.cpp:90: Failure
Value of: testing::Mock::VerifyAndClearExpectations(mctpEp.get())
  Actual: false
Expected: true

[  FAILED  ] NVMeRecovery.optimisationFailure (8348 ms)
[----------] 1 test from NVMeRecovery (8359 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (8415 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] NVMeRecovery.optimisationFailure

 1 FAILED TEST
==197574==
==197574== HEAP SUMMARY:
==197574==     in use at exit: 8 bytes in 1 blocks
==197574==   total heap usage: 7,500 allocs, 7,499 frees, 1,751,392 bytes allocated
==197574==
==197574== 8 bytes in 1 blocks are still reachable in loss record 1 of 1
==197574==    at 0x4840F83: operator new(unsigned long) (in /usr/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
==197574==    by 0xD157EF: testing::internal::GetFailureReporter() (gmock-internal-utils.cc:124)
==197574==    by 0x9C16E3: testing::internal::Expect(bool, char const*, int, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) (gmock-internal-utils.h:259)
==197574==    by 0x9C2303: testing::internal::UntypedFunctionMockerBase::FailureCleanupHandler::~FailureCleanupHandler() (gmock-spec-builders.h:1427)
==197574==    by 0x9D852C: testing::internal::FunctionMocker<void ()>::InvokeWith(std::tuple<>&&) (gmock-spec-builders.h:1898)
==197574==    by 0x9CB22B: testing::internal::FunctionMocker<void ()>::Invoke() (gmock-spec-builders.h:1548)
==197574==    by 0x9C2EF8: MockMctpEndpoint::recover() (test_nvme_mi_recovery.cpp:27)
==197574==    by 0xAD57EB: NVMeMi::recover() (NVMeMi.cpp:192)
==197574==    by 0xAD50B1: NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}::operator()(std::error_code const&) const (NVMeMi.cpp:167)
==197574==    by 0xAFF26F: void std::__invoke_impl<void, NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}&, std::error_code const&>(std::__invoke_other, NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}&, std::error_code const&) (invoke.h:61)
==197574==    by 0xAF3ED2: std::enable_if<is_invocable_r_v<void, NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}&, std::error_code const&>, void>::type std::__invoke_r<void, NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}&, std::error_code const&>(NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}&, std::error_code const&) (invoke.h:111)
==197574==    by 0xAEFE8F: std::_Function_handler<void (std::error_code const&), NVMeMi::epOptimize()::{lambda(boost::system::error_code)#1}::operator()(boost::system::error_code) const::{lambda(std::error_code const&)#1}>::_M_invoke(std::_Any_data const&, std::error_code const&) (std_function.h:290)
==197574==
==197574== LEAK SUMMARY:
==197574==    definitely lost: 0 bytes in 0 blocks
==197574==    indirectly lost: 0 bytes in 0 blocks
==197574==      possibly lost: 0 bytes in 0 blocks
==197574==    still reachable: 8 bytes in 1 blocks
==197574==         suppressed: 0 bytes in 0 blocks
==197574==
==197574== For lists of detected and suppressed errors, rerun with: -s
==197574== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
```

Change-Id: Idf3fb6d167a9e11ea1d554c4ab2f2c50d1489421
Signed-off-by: Andrew Jeffery <andrew@codeconstruct.com.au>