Agent resource cache updates

[LIU Yulong]

1. Agent resource cache has an infinite growth set: _satisfied_server_queries
https://github.com/openstack/neutron/blob/master/neutron/agent/resource_cache.py#L41

there is no entry removal for this set.

2. Because this set has a non-standard structure, for instance:
set([('Port', ('id', (u'830d035e-5138-49ae-bbe4-324f4096656d',))), ('Network', ('id', ('e04208de-1006-4a6b-881a-83129856afa6',))), ('Network', ('id', ('505155ea-8fbb-42d1-a8c9-cc2c78f8476e',))), ('Port', ('id', (u'ac825cc9-906a-45db-a77d-4e336fc1c4ea',))), ('Port', ('id', (u'c3a72a39-dbd5-4737-a68c-120de93b186c',))), ('Network', ('id', ('cd5155df-9777-4487-a730-b5ee533c4f80',))), ('Port', ('id', (u'340e02f2-fe54-4f31-8858-7f6413fb0010',))), ('Port', ('id', (u'64fc4d85-04d6-453f-8d20-f4d1308d34fd',))), ('Network', ('id', ('a6201723-a237-433c-b357-82a6a24526e5',))), ('Network', ('id', ('71a80697-1705-4bd0-b65b-0fd7dd616836',))), ('Port', ('security_group_ids', ('48a2ebb8-16ea-4a0a-9d45-eabc6a6b3dcf',))), ('Network', ('id', ('7e83c48a-b246-4a02-bb87-10016ac4b47e',))), ('SecurityGroupRule', ('security_group_id', (u'48a2ebb8-16ea-4a0a-9d45-eabc6a6b3dcf',))), ('Port', ('id', (u'2cc656ba-b07b-4e85-ad56-ee6da4b2e763',))), ('Port', ('id', (u'89d0aab8-82f7-4e5e-98b1-e009e31498ce',))), ('Port', ('id', (u'd820dbc2-bf4f-463b-9a67-6b704202bee0',))), ('Network', ('id', ('aea5771b-9655-4936-b9f4-f94d482c0b15',))), ('Port', ('id', (u'68c3e31b-9bf6-45e9-bfbb-3da1cafebcec',)))])

It's hardly to remove all entries for one resource at all, because if some codes query cache by filter=None, some codes use fitler={"x":y, "a": b}, the entries are various, especially when the code is not in Neutron.

3. If the port removed, and added again, because the query is in the _satisfied_server_queries: https://github.com/openstack/neutron/blob/master/neutron/agent/resource_cache.py#L75

It may return None or stale resource.

4. If users query one resource like:
q1 = {"id": 1}, and return R
q1 = {"id": 1, name="2"}, and return None
q1 = {"id": 1, device="3"}, and return None

The set will also grow up. And it does not avoid bulk_pull call to neutron-server at all.

So, it's better to remove such "server_queries" records. Because, if the resource is in cache, just return it. If it is not, get it from neutron-server.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/neutron/+/904169