Reset HttpCacheSM on following redirect

[masaori335]

We face an issue that http.current_active_client_connections metrics keep increasing. After investigation with @cmcfarlen and @moonchen, we found HttpSM hangs on state_cache_open_read when it follows redirect and HttpCacheSM::captive_action was cancelled. Because HttpCacheSM::state_cache_open_read does nothing.

trafficserver/src/proxy/http/HttpCacheSM.cc

Lines 106 to 112 in b1e9327

	STATE_ENTER(&HttpCacheSM::state_cache_open_read, event);
	ink_assert(captive_action.cancelled == 0);
	pending_action = nullptr;
	if (captive_action.cancelled == 1) {
	return VC_EVENT_CONT; // SM gave up on us
	}

To fix the issue, this PR does reset captive_action.cancelled and other counters of HttpCacheSM on following redirect.

From our testing, this patch seems working, however, captive_action and pending_action related code seems pretty odd. They should be fixed.

[JosiahWI]

Would also be interesting to know whether this affects #11700.

[JosiahWI]

Very nice fix. I believe I found this bug during testing earlier, but I did not have time to confirm it or track down what caused it.

I agree we need to look at how captive_action and pending_action are handled... I think other bugs are probably also related to it.

[masaori335]

For the record, the issue happens only on 10.0.x and master. Because we reverted commits from 9.2.x branch.
#9338

[cmcfarlen]

Cherry-picked to 10.1.x branch