Issue 19 one bond per user

[wjwwood]

This is an alternative implementation to issue #19.

It uses one bond per client instead of one bond per capability "use".

@bit-pirate for review

[wjwwood]

@bit-pirate bump on the review

I think this pull request is superior to #19.

The main difference is that #19 establishes one bond per call to /use_capability, whereas this pr establishes one bond per client to the capability server.

19 work flow:

• For each capability:
  • Call ~use_capability
  • Establish bond
  • Later call ~free_capability or break bond

This pull requests work flow:

• Call ~establish_bond
• Establish the bond
• For each capability:
  • Call ~use_capability
  • Later call ~free_capability
• Break bond to free any remaining uses which were not expilicitly free'ed

The entire work flow is hidden by the client API, for both pull requests, this is how the client would use the capabilities.client Python API:

from capabilities.client import Client
client = Client()
# Use the line below if the capability_server has a different name
# client = Client(capability_server_node_name='/capability_server_node_name')
client.wait_for_services()  # Optional
client.use_capability('foo_pkg/Foo')
client.use_capability('foo_pkg/Foo')
client.use_capability('bar_pkg/Bar')
client.free_capability('foo_pkg/Foo')
client.shutdown()

[bit-pirate]

PR #50 fixes the two issues mentioned above. Apart from that this PR looks really nice. I tested freeing/using capabilities and corner cases, such as killing the app manager after an app depending on caps has been started, and it works well (caps get freed correctly)! :+1: