Have a debugging story for credentials

[atheriel]

Now that we do so much magic internally to get ambient credentials, I wonder if we need better ways to surface (and/or debug) what credentials have been selected to users.

Something like the following:

> chat <- chat_snowflake()
## Using model = "llama3.1-70b".
## Using key-pair authentication.

or

> chat <- chat_databricks()
## Using machine-to-machine (M2M) credentials.

or

> chat <- chat_azure()
## Using Azure service principal credentials.

Of course, that might be annoying if we print it every time; I know that gargle at least has a "verbosity" setting that turns on debug messages for the credential source. The internal API could be:

set_auth_source <- function(src) {
  if (!getOption("ellmer_auth_debug", FALSE)) {
    return()
  }
  cli::cli_inform("Using {.field {src}}.")
}

[atheriel]

Note that this won't work all that well for cases where we outsource finding credentials, e.g. with paws.common.

[hadley]

I wonder if we should consider an optional dependency on logger (or log4r) so we could log this info to a stream that the user could optionally enabled.

[mladencucak]

I would support some kind of logging...never properly used in many years of R but now log every step in Python because many things can go wrong as all providers constantly change things and there is a lot of instability on the provider side.