Fixed data access connections to subprograms

[AaronGreenhouse]

Fixes #2161

[stevevestal]

A couple of general comments that might be relevant to long-term plans for instantiating end-to-end connections.

The earlier convention was to have instantiated connections go from thread port to thread port, but not dive down into thread internals. Tool developers had to do their own stuff to get internal details as needed (which we are revisiting in the case of access connections to data sub-components). Many kinds of connections can do this, not just data access connections. Going forward, if the plan is to move in general towards instantiating connections from true first-feature to last-feature, something to consider is the impact on legacy tools from a variety of third-party developers.

In such cases, it is nice to be able to easily determine the containing thread and the feature of the containing thread through which a connection passes (e.g., useful to obtain properties such as processor bindings that do not apply to data features and thus are not inheritable but might be relevant to some analyses); and to easily determine the containing process and the feature of the containing process through which a connection passes (e.g., the AADL ARINC 653 Annex has places where it says, or at least strongly implies, that properties of the process rather than the thread ports specify certain things about connections between A653 partitions).

[lwrage]

@stevevestal Currently we have no plans to change connection instantiation further. Regarding the impact on existing tools: If our change breaks your tools, let's discuss via the google group.