Fix Metropolis for sparse distributions

[stavros11]

Fixes #346 using the idea mentioned there. Execution time and KL divergence for random distributions are the same as our existing code:

nshots	master time (sec)	fix time (sec)	KL master	KL fix
100000000	0.148165	0.238486	1.99e-05	1.25e-05
200000000	0.309631	0.478443	1.15e-05	9.90e-06
400000000	0.458794	0.828578	1.23e-05	8.05e-06
600000000	0.676857	1.156601	9.35e-06	8.28e-06
800000000	1.037136	1.502599	1.14e-05	6.70e-06
1000000000	0.986530	1.843876	1.03e-05	8.21e-06
2000000000	1.920199	3.533490	9.72e-06	7.11e-06
4000000000	3.620555	7.008676	9.70e-06	6.89e-06
6000000000	5.468573	10.519186	9.39e-06	7.12e-06
100	0.001501	0.001475	1.80e+01	1.82e+01
10000	0.002878	0.002937	1.51e-01	1.57e-01
10000000	0.022121	0.027112	1.16e-04	9.69e-05

nqubits	master time (sec)	fix time (sec)	KL master	KL fix
6	0.020047	0.036727	1.57e-05	8.39e-06
8	0.023921	0.029751	3.62e-05	2.38e-05
10	0.022121	0.027112	1.16e-04	9.69e-05
12	0.026822	0.034296	5.29e-04	3.59e-04
14	0.024820	0.040488	1.75e-03	1.56e-03
16	0.032375	0.042690	7.62e-03	6.16e-03
18	0.037547	0.060351	3.12e-02	2.92e-02
20	0.083534	0.113640	1.96e-01	1.72e-01
22	0.271751	0.321594	2.22e+00	1.98e+00
24	0.891320	0.934847	6.75e+00	6.53e+00
26	3.006421	3.102790	8.49e+00	8.42e+00

[codecov]

Codecov Report

Merging #347 (a3ea890) into master (4ed77c7) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master      #347   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           76        76           
  Lines        12387     12406   +19     
=========================================
+ Hits         12387     12406   +19     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...qibo/tests_new/test_tensorflow_custom_operators.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4ed77c7...a3ea890. Read the comment docs.

[scarrazza]

Thanks looks good.