Node ID should be available in RPC errors

[dfawley]

1. Include node ID with all xdsclient errors possible

2. Audit all possible ways an xds-enabled gRPC channel can fail as a result of xdsclient problems, and ensure the xdsclient error is propagated to RPC errors

[dfawley]

@easwars can you help scope this, please?

[easwars]

I suggest making the following changes:

• The watch resource API call on the xDS client can fail for a bunch of reasons. We need to include the node ID in all these errors:
  • A resource type implementation was already registered.
  • Authority in the resource name is not found in the bootstrap configuration.
  • Creation of transport to the management server fails.
• A resource is NACKed:
  • The OnError callback needs to include the xDS node ID in the returned error.
  • The error is also stored in the cache to return to future watchers. This should include the xDS node ID as well.
• Reporting connectivity failures from xDS client to watchers. xDS node id should be part of these errors.
• CDS lb policy needs to include the xDS node id in errors
  • Requested resource is not found
  • Sends a resolver error to child policy or puts the channel in TF
• xDS resolver needs to include the xDS node id in the following cases
  • Error reported to the channel (when requested resource is not found)
  • Error reported from the config selector
• xDS server needs to include the xDS node id when it puts the server in non-serving mode
• All resource watchers' OnResourceDoesNotExist callback should include the xDS node id if they are throwing a log. Also, these logs should not be Info protected with a verbosity check, but should be Error logs since resource not found is usually something that is worth sharing with the user by default.