One of the first things I do on most clusters is set the idle timeouts to avoid having many open queries from clients that hold queries open (Hue for example). It would be nice to have these set out of the box to reasonable values to avoid long hanging queries.
Personally I set to between 15 mins - 1 hour. 1 hour seems to be a reasonable high side default IMHO.
Setting a reasonably high default idle_query_timeout is unlikely to disrupt well-behaved workloads because it only kicks in when a query is consuming resources and the client is not interacting with it in any way. Some clients (like Hue) do keep queries open for a long time between fetches, e.g. if users page through results slowly, but in practice it's rarely, if ever, desired to keep queries open in this case.
Setting a default idle_session_timeout would be more disruptive because clients may keep connections open for various legitimate reasons even if they're not running queries, e.g. if they are caching connections. This is fairly innocuous because sessions with no running queries do not consume excessive resources.
Setting idle_session_timeout is riskier because there are more reasons why clients would keep a connection open, even if not running queries. Running Impala tests with a 15 minute session timeout caused a lot of query failures, because various connections used by longer-running tests like test_ddl were only used intermittently or at the beginning and end of tests.