The binary fails if the fibril is synced implicitly over its
destructor. It appears that after membar(doSync()), the base pointer
is wrong, causing the subsequent "reverseStealCount = 0" in
syncWaitFree() to write to unallocate memory. I am not sure yet why,
but this works around the issue.

This also refactors a bit when we take timestamps.

We do not need to measure the work duration which is the known pulse.
This makes comparing the raw data easier because the latency does not
include the pulse duration.

The default is Burst mode which inserts new work as fast as possible
sleeping until the next pulse.
Even mode distributes the new work evenly in the duration between two
pulses.

This allows emper to be build with --meson-fatal-warnings on in our i4lab.

Fixes builds on Debian stable, which currently fail with

ccache c++ -Ieval/io_latency.p -Ieval -I../eval -Iemper -I../emper -Iemper/io -I../emper/io -Iemper/include -I../emper/include -I/usr/include -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wnon-virtual-dtor -Wextra -Wpedantic -Werror -std=c++2a -O2 -g -DBOOST_ALL_NO_LIB -pthread -isystem../subprojects/liburing-liburing-2.0/src/include -isystemsubprojects/liburing-liburing-2.0/src/include -MD -MQ eval/io_latency.p/IoLatency.cpp.o -MF eval/io_latency.p/IoLatency.cpp.o.d -o eval/io_latency.p/IoLatency.cpp.o -c ../eval/IoLatency.cpp../eval/IoLatency.cpp: In function ‘void watchFunc()’:
../eval/IoLatency.cpp:45:17: error: ‘struct std::atomic<bool>’ has no member named ‘wait’
   45 |   readSubmitted.wait(false, std::memory_order_acquire);
      |                 ^~~~

* increase iterations to 10K this produced more stable averages
* add options to dump all measurements
* add baseline benchmark using two threads and read/write syscalls
* add iouring benchmark using two threads and iouring to read

It uses a watch thread writing to eventfd and a ReadFuture with a
Callback to measure the latency between writing to the eventfd and
executing the resulting IO Continuation.

The worker submitting the read request starts a busy loop after the
submission to enforce that the IO-continuation is delayed on this
worker.

While we do not have yet LLVM 13 in the gitlab-ci, I use it on my
systems. So fix the new warnings found with clang-tidy 13.

The emper header LockedUnboundedQueue.hpp could depend on different libraries
according to the implementation.

To link those dependencies with everything including LockedUnboundedQueue.hpp
we propagate all emper_dependencies through emper_dep.

And using emper_dep as a dependency seems anyway better than essentially
writing down emper_dep manually each time.

emper_dep essentially is:
(link_with:emper, include_directories: emper_all_include)

This introduces AbstractWorkStealingScheduler which holds the common
work-stealing scheduling strategy.

In debug builds, the schedule() method is now guarded by an assert(),
as due to this change, the schedule() method will no longer work from
everywhere.

This also improves the worker sleep method.

And fix the #include issues reported by the newer IWYU version.

It's 'ns' stupid.