remove unused memoisation

[rade]

Remove all memoisation and then re-introduce it a) for top-level renderes, i.e. those closest to the API, and b) shared renderes.

This cuts the number of memoising renderers from 74 to 21.

I ran a number of tests with a recent prod report

prog/scope --mode=app --weave=false --app.collector=file:report.json

and then timed the fetching of topologies with

time curl -so /dev/null http://localhost:4040/api/topology/<topology>

topology	befor	rpeat	ahost	hits	after	rpeat	ahost	hits
processes	0.315	0.108	0.125	0/1/1	0.312	0.108	0.124	0/1/1
processes-by-name	0.249	0.029		0/1	0.229	0.026		0/1
containers	1.282	0.399	0.421	0/1/1	1.276	0.416	0.432	0/1/1
containers-by-hostname	0.926	0.053		1/1	0.906	0.058		1/1
containers-by-image	0.924	0.034		0/1	0.898	0.038		0/1
pods	0.954	0.074	0.082	1/1/1	0.960	0.080	0.089	2/1/1
kube-controllers	0.947	0.042		1/1	0.943	0.039		1/1
services	0.960	0.031		1/1	0.936	0.031		1/1
hosts	1.278	0.025		4/1	1.237	0.025		4 /1

rpeat = fetching the same topology again
ahost = fetching the topology after first fetching the host topology
hits = memoisation cache hits during fetch / rpeat fetch / ahost fetch

Timings are best run out of five.

So there really is no difference in timing, but it should consume less memory.

Each of the memoisations does have an effect, often very substantial; i.e. disabling any one increases at least one of the numbers above.

Fixes #2923