Log exception class+message on token-read errors, traceback at DEBUG

Address review feedback from @bdrx312 on PR #69074: the original
`log.warning` calls in the SaltDeserializationError and OSError
handlers in `LoadAuth.get_tok` did not capture the exception at all,
making troubleshooting harder. Two combined changes:

1. Add the exception class+message inline via `%r` so each
   occurrence is one greppable line that names which subclass
   (e.g. `ConnectionResetError`) and its message produced the
   warning -- enough to triage without a full traceback.
2. Add a companion `log.debug(..., exc_info=True)` per @bdrx312's
   suggestion. Operators who need the full traceback for a specific
   intermittent failure can raise the level to DEBUG and see it.
   Costs nothing at the default WARNING level because logging skips
   exc_info formatting when the level is disabled.

This avoids GBs/hour of stack frames during a flapping Redis or NFS
outage (the original `exc_info=True` approach considered earlier)
while keeping full diagnostic depth one log-level change away.

Refs: #69073

Author: co-cy

salt/auth/__init__.py

@@ -243,31 +243,43 @@ def get_tok(self, tok):
             tdata = self.tokens["{}.get_token".format(self.opts["eauth_tokens"])](
                 self.opts, tok
             )
-        except salt.exceptions.SaltDeserializationError:
+        except salt.exceptions.SaltDeserializationError as exc:
             # The on-disk / in-store token blob is corrupt and cannot
             # be parsed. Removing it is the right call -- a corrupt
             # token can never authenticate anyway, and leaving it
             # around makes every subsequent ``get_tok`` for the same
-            # id keep failing.
+            # id keep failing. ``%r`` on the exception gives the
+            # operator the class and message inline (e.g. msgpack
+            # format error, truncated file) without spamming a full
+            # traceback into a hot-path WARNING; the full traceback is
+            # available via the companion ``log.debug`` for deeper
+            # investigation.
             log.warning(
-                "Token %r could not be deserialized; removing it from the store.",
+                "Token %r could not be deserialized (%r); removing it from the store.",
                 tok,
+                exc,
             )
+            log.debug("Token deserialization traceback:", exc_info=True)
             self.rm_token(tok)
             return {}
-        except OSError:
+        except OSError as exc:
             # Transient backend error (Redis connection blip, NFS hang,
             # hung disk). The token itself is fine; do NOT delete it --
             # that would log every authenticated user out on every
             # backend hiccup. Return an empty dict so the caller treats
             # this request as not-authenticated; the next request will
             # retry against the backend and succeed once it recovers.
+            # Same logging pattern as above -- exception class + message
+            # at WARNING, full traceback at DEBUG so a flapping deploy
+            # stays diagnoseable without GB/hour of stack frames.
             log.warning(
-                "Token store transient error reading %r; treating as "
+                "Token store transient error reading %r (%r); treating as "
                 "not-authenticated for this request without removing the "
                 "token from the store.",
                 tok,
+                exc,
             )
+            log.debug("Token store transient-error traceback:", exc_info=True)
             return {}
 
         if not tdata: