Reject config with unknown directives before committing to it

[rousskov]

Ideally, we want to reject configurations with unknown directives before
applying any configuration changes that correspond to known directives,
but current apply-as-you-parse architecture makes that impractical.
Pending smooth reconfiguration refactoring will make that possible, but
we can make a step towards that ideal future now.

Rejecting bad configurations before calling configDoConfigure() reduces
the set of configuration errors that Squid can detect in one execution
(because configDoConfigure() error-checking code is not reached), but
that small reduction is a lesser evil compared to running
configDoConfigure() with a clearly broken config, especially when we are
going to kill Squid anyway. While many legacy parse_foo() functions do
apply significant changes before configDoConfigure(), we cannot easily
prevent that (for now). We can easily prevent configDoConfigure().

[rousskov]

This PR is a yet another tiny step towards smooth reconfiguration support, addressing an existing TODO. It is probably the last change I can carve out before #1840 is finally merged. Isolating and maintaining these pending changes wastes a lot of my time. Please move #1840 forward!

This review annotates this PR but does not request any changes.

[rousskov]

It may be tempting to throw even earlier, when unrecognizedDirectives constant is set above defaults_if_none() and defaults_postscriptum() calls. However, we should not do that. It is not obvious in the current code, but those two functions can also detect unrecognized directives and similar problems. Pending smooth reconfiguration changes will make this code look natural, removing the current "Why not throw earlier?" question.

[kinkie]

LGTM