Adjust RouteCache to mitigate contention

[FelixMcFelix]

This PR follows up on the steps outlined in #997. The main changes are to increase the table size, and to minimise the amount of time holding the write lock on a miss by fetching a new route before taking the lock. This includes refusing to attempt an insert on a failed lookup in a full table.

One of the steps outlined in the issue, not evicting present entries, was already implemented.

Closes #997.

[FelixMcFelix]

Some quick testing notes on throughput on glasgow during the implementation of the above:

	-P 1	-P 64	-P 128
master	14.0	31.3	31.3
MAX_CACHE_ENTRIES = 32	14.2	27.4	25.9
+ reduced lock scope	14.0	29.1	28.7
+MAX_CACHE_ENTRIES = 8192	14.2	31.7	31.7

My read is that the lock scoping changes (and refusal to take a write at suspected max capacity) do help in the cases that we have more flows than slots. I'll see about getting some lockstat runs.