Import is slow

[dabacon]

import cirq takes around 1.5 seconds (this is in colab).

Would be nice to get some other numbers from users as to what they are seeing.

[cduck]

New MacBook, Python 3.8, Cirq master (7191cb57):

$ python -c 'import time; time.sleep(1); t1 = time.monotonic(); import cirq; t2 = time.monotonic(); print(t2-t1)'
1.65165978 (±0.05)

[vtomole]

Ubuntu 18.04

(venv3.6) vtomole@vtomole:~$ python -c 'import time; time.sleep(1); t1 = time.monotonic(); import cirq; t2 = time.monotonic(); print(t2-t1)'
4.911438770999666

[alexandrupaler]

Ubuntu 18.04. i5-8250U
python3 -c 'import time; time.sleep(1); t1 = time.monotonic(); import cirq; t2 = time.monotonic(); print(t2-t1)' executed sequentially

2.7808562550017086
1.722753725000075
1.588053849998687
1.700277088002622
2.827545382999233

[cduck]

Cirq submodule import times excluding external packages:

Worst load times:
0.569  cirq
0.199  cirq.google
0.137  cirq.qis
0.136  cirq.qis.measures
0.131  cirq.linalg
0.130  cirq.linalg.decompositions
0.126  cirq.google.engine
0.126  cirq.google.engine.engine
0.122  cirq.google.engine.engine_client
0.080  cirq.google.engine.client.quantum
0.080  cirq.google.engine.client.quantum_v1alpha1
0.059  cirq.google.engine.client.quantum_v1alpha1.gapic.quantum_engine_service_client
0.038  cirq.google.devices
0.038  cirq.google.devices.known_devices
0.026  cirq.google.api

[dabacon]

Using tuna, the top offenders are numpy, sympy, matplotlib, pandas, and networkx. We've added code #4653 to do LazyLoading for scipy which was also slow. The good news is that the problem is mostly these external deps, not cirq.