Extend TypeDescription codegen with referenced type inclusion, disable option, raw sources

[emersonknapp]

Part of ros2/ros2#1159

Bundled with (must be merged together):

• Type Description Nested Support ros2/rosidl_typesupport#141
• Type Description Nested Support ros2/rosidl_typesupport_fastrtps#101
• Use get_type_hash_func on typesupports (rep2011) ros2/rmw_fastrtps#680
• Use get_type_hash_func on typesupports (rep2011) ros2/rmw_connextdds#113
• Use get_type_hash_func on typesupports (rep2011) ros2/rmw_cyclonedds#448
• Use get_type_hash_func on typesupports (rep2011) ros2/rcl#1053

Replaces (or extends) ros2#727. Key differences

• type_support_t->get_type_hash_func rather than bare value
• get_type_description and get_type_sources take a typesupport struct argument (for use by dynamic types)
• Type descriptions and sources include referenced types, rather than baking in full recursive dependencies per-interface
• Expected hashes of referenced types are included in codegen to optionally error-check with at runtime

Key features:

• Changes TYPE_HASH_TUPLES to more accurate TYPE_DESCRIPTION_TUPLES in cmake
• Adds new C interface functions get_type_description, get_individual_type_description_source, and get_type_description_sources
• Implements above, generating code for type descriptions and raw interface definition sources, embedded in the binaries
• Optionally disables embedding that information in the generated code by the CMake flag ROSIDL_GENERATOR_C_DISABLE_TYPE_DESCRIPTION_CODEGEN
• Refactors rosidl_generator_type_description to output a new structure of type_description_msg plus a flat list of all referenced types and their hashes at the time of generation. These hashes are used at runtime (in Debug builds) to detect ABI breaks when constructing type descriptions the first time

[emersonknapp]

Windows also needs to open(encoding='utf-8') explicitly:

• Windows

[emersonknapp]

Note for reviewers - these files are the copied generated sources, so they give a good sense for what the templates do, since those are not super easy to read

[wjwwood]

lgtm

[clalancette]

In short, I don't have much to say here; it looks generally good. I'll suggest we go ahead and merge this one into ros2#727 now, and run CI. I'll do one final pass on the combination before approving over there.