* [iron] Bugfix for writer not being able to open again after closing (backport #1599) (#1635)

* re-applies fixes from #1590 to rolling. Also removes new message definition in sequential writer test for multiple open operations. Also clears topic_names_to_message_definitions_ and handles message_definitions_s underlying container similarly. Lastly, also avoids reset of factory in the compression writer, adds unit test there too.

Signed-off-by: Yannick Schulz <yschulz854@gmail.com>
Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* removes unused compressor_ member from compresser writer class. Also delegates rest of the closing behavior to the base class in close method, as it is handled in the open and write methods of the compression writer

Signed-off-by: Yannick Schulz <yschulz854@gmail.com>

* Remove unrelated delta

- message_definitions_ was intentionally allocated on the stack and
should persist between writer close() and open() because it represents
cache for message definitions which is not changes.

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Don't call virtual methods from destructors

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Cleanup 'rosbag2_directory_next' after the test run

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Protect Writer::open(..) and Writer::close() with mutex on upper level

- Rationale: To avoid race conditions if open(..) and close() could be
ever be called from different threads.

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Bugfix for WRITE_SPLIT callback not called for the last compressed file

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Bugfix for lost messages from cache when closing compression writer

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Address build failure by using rcpputils::fs instead of std::filesystem

- Note: On Iron we haven't migrated to the std::filesystem and using
rcpputils::fs

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Adopt failing 'open_succeeds_twice' test for Iron

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Return from writer's open() immediately if storage already open

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

---------

Signed-off-by: Yannick Schulz <yschulz854@gmail.com>
Signed-off-by: Michael Orlov <michael.orlov@apex.ai>
Co-authored-by: Yannick Schulz <yschulz854@gmail.com>
(cherry picked from commit a360d9b97e)

# Conflicts:
#	rosbag2_compression/src/rosbag2_compression/sequential_compression_writer.cpp
#	rosbag2_cpp/src/rosbag2_cpp/writer.cpp
#	rosbag2_cpp/src/rosbag2_cpp/writers/sequential_writer.cpp

* Address merge conflicts

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Fix for segfault in open_twice test

- Ensure that writer_ is destructed before intercepted fake_metadata_

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Fix for "open_succeeds_twice" test failure on second run

- Use std::filesystem for temp files and folders operation. For some
reason rcpputils::fs::delete_all(folder_name) wasn't able to delete temp
folder with subfolders.

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

* Adopt changes in TestRosbag2CPPAPI::minimal_writer_example for humble

- The `serialized_msg2` is not owning the serialized data after the
first call writer.write(serialized_msg2,..). i.e. need to use another
message or another API in test. This is not a bug - this is by design.

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>

---------

Signed-off-by: Michael Orlov <michael.orlov@apex.ai>
Co-authored-by: Michael Orlov <michael.orlov@apex.ai>