PR6878 (-cc1 should default to -fexceptions)

[Rafael Espíndola]

> And then the driver basically makes the same decision that it does today,
> and if that differs from what it knows the frontend will decide, it should
> pass the appropriate option.

The -cc1 interface is not normally used by humans (other than us when
writing tests), so why is this mild duplication better? It seems that
if the driver needs to reason about some options, it should just pass
the result to -cc1. If the options don't change the driver behavior,
it should probably just forward them.

Cheers,
Rafael