Package redact implements functions to redact sensitive information from
errors. A basic example is:

	name := "Alex"
	id := 5
	err := redact.Errorf("error getting user %s with ID %d",
		name, Safe(id))

	fmt.Println(err)
	// error getting user Alex with ID 5

	fmt.Println(redact.Error(err))
	// error getting user <redacted string> with ID 5

See the package docs and tests for more examples.

This will allow us to get more information in Sentry without sending any
identifying information.

Make `safeError` satisfy `fmt.Format` so that it can print stack traces
the same way pkgs/errors does. Don't double redact errors so that
already-redacted output isn't replaced with a placeholder.

Clean up and fix telemetry so that it can be actually useful. This
commit is rather large because telemetry happens to touch a lot of
different things.

Example of what it looks like in Sentry: https://jetpack-io.sentry.io/issues/4047374127

Client
------

- Remove unnecessary telemetry configuration, such as `InitOpts`. Make
  the `build` package the source of truth for version information and
  link-time variables. Generate a global execution ID once at startup.
- Standardize the contexts to use names and values recognized by Sentry.
- Remove arbitrary CLI args and DEVBOX_ environment variables from the
  context, as these were leaking information.

Errors
------

- In combination with the `redact` package, errors can now include data
  that has been marked as safe for telemetry.
- Errors that haven't been marked as safe have a more helpful
  placeholder containing the chain of error types up until the first
  exported error.

Stack Traces
------------

- Use the correct file and function names. This allows Sentry to group
  errors correctly.
- Don't record every wrapped error. This doesn't make sense in Go since
  an error is a value containing the entire message. Otherwise you end
  up with a bunch of error messages that repeat themselves.