Starting to address #6. This is just the start of the work, but it gives
us something to build on.

Basically, what we're trying to do is build log sink functionality into
terraform-plugin-log. Normally, Terraform acts as a log sink, gathering
up logs from core and from all the plugins, filtering them
appropriately, and combining them into a single log output stream, and
deciding where to send that stream. However, when running acceptance
tests, the plugin is actually the longest-running process, and so this
model breaks down. Instead, we need to have the plugin act as the sink,
gathering the logs from core and the plugin and other plugins, combining
them into a single stream, filtering the stream, and deciding where to
output the stream.

Rather than implement this functionality multiple times in different
SDKs, it makes more sense to add it to the tfsdklog package and just
call it from the test frameworks.

There are a few parts to this. First, we need a new sink logger that all
other loggers are derived from. It should read environment variables to
determine what level of output is desired, what format of output is
desired, and where to send that output. Second, we should make all our
loggers be sub-loggers of that sink when the sink is set. Third, we
should make sure our sub-loggers can only log at levels equal to or
higher than the level of this new sink logger when it's set. And
finally, we should provide functionality to process hclog JSON output
(like from terraform when TF_LOG=json) and route it through this sink
logger, so we can combine the log output of the CLI and other providers
with the log output native to the process.

This commit falls short in a few places:

* sub-loggers aren't limited to equal-or-higher levels from the sink
  logger yet.
* none of this has been tested, even manually. It builds?
* we need to check that all the environment variables we need to be
  honoring are getting honored for the provider under test, the CLI, and
  the providers the CLI launches.

This depends on #10.