Algorithm Nodes should NOT cache outputs by default

[mpu-creare]

Description
This is a big gotcha for new users. When developing a new algorithm node, if they (1) evaluate the results, (2) then change the code, and (3) evaluate the results again, nothing will change because the output is retrieved from the cache.

Steps to Reproduce

class MyAlg(Algorithm):
     data = NodeTrait()
     def algorithm(self, inputs):
         return inputs['data']
my_alg = MyAlg()
o = my_alg.eval(coords)

class MyAlg(Algorithm):
     data = NodeTrait()
     def algorithm(self, inputs):
         return inputs['data'] * 0 
my_alg = MyAlg()
o2 = my_alg.eval(coords)

assert np.all(o2.data != o.data)

This throws an assertion error .

Expected Behavior
I would expect o2 to be all zeros.

Observed Behavior
o2 == o, since it comes from the cache.

Additional Notes
There are multiple ways we could solve this problem. I think the best way is to make cache_update True by default for Algorithm and Compositor Nodes, and then in the documentation note how to optimize a pipeline that uses the same Algorithm/Compositor node in multiple places.

[jmilloy]

I think we are going to set cache_outputs = tl.Bool(default_value=False) for Algorithm nodes, maybe for other nodes as well. The caching is an optimization that some user's will want to turn on.