Implemented parallel tiling

[micheles]

Closes #8878. Here is the performance on the spot machine for EUR with 1% of the sources:

# master, 6 tiles
| calc_2, maxmem=223.0 GB    | time_sec | memory_mb | counts     |
|----------------------------+----------+-----------+------------|
| total classical            | 57_831   | 10.2      | 24_882     |
| get_poes                   | 32_622   | 0.0       | 28_062_228 |
| computing mean_std         | 14_299   | 0.0       | 262_240    |
| composing pnes             | 7_965    | 0.0       | 28_062_228 |
| total postclassical        | 7_113    | 3_365     | 86         |
| combine pmaps              | 4_661    | 0.0       | 176_738    |
| total preclassical         | 2_084    | 151.2     | 616        |
| ClassicalCalculator.run    | 1_724    | 5_931     | 1          |
| read PoEs                  | 1_040    | 3_365     | 86         |

# parallel tiling
| calc_3, maxmem=232.6 GB    | time_sec | memory_mb | counts    |
|----------------------------+----------+-----------+-----------|
| total classical            | 38_351   | 3.44085   | 5_856     |
| get_poes                   | 22_403   | 0.0       | 8_064_162 |
| total postclassical        | 8_586    | 3_163     | 86        |
| composing pnes             | 7_048    | 0.0       | 8_064_162 |
| computing mean_std         | 6_958    | 0.0       | 60_315    |
| combine pmaps              | 4_164    | 0.0       | 176_738   |
| read PoEs                  | 3_422    | 3_137     | 86        |
| total preclassical         | 2_046    | 148.7     | 616       |
| ClassicalCalculator.run    | 1_262    | 17_292    | 1         |

The performance is much better, as expected (21 minutes instead of 29 minutes overall) and the improvement is even better on an HPC (expected more than 2x overall). read PoEs becomes 3x worse, but it is a minor issue at the moment (in this example only 68.5 GB of PoEs are stored).
The task distribution is much better:

| operation-duration | counts | mean    | stddev | min     | max     | slowfac |
|--------------------+--------+---------+--------+---------+---------+---------|
| classical          | 2_088  | 27.7    | 44%    | 0.13246 | 58.9    | 2.12525 |
| classical          | 298    | 128.7   | 36%    | 0.38588 | 185.4   | 1.44099 |

The data transfer it much better too:

| task              | sent                                         | received  |
|-------------------+----------------------------------------------+-----------|
| classical         | sitecol=4.67 GB cmaker=11.65 MB srcs=7.34 MB | 81.17 GB  |
| classical         | sitecol=3.82 GB srcs=59.47 MB cmaker=9.98 MB | 48.85 GB  |

For small calculations the performance is expected to be much worse, I have seen even 3x worse in the test share_small.ini:

# master
| calc_69835, maxmem=5.3 GB | time_sec | memory_mb | counts  |
|---------------------------+----------+-----------+---------|
| total classical           | 492.6    | 114.8     | 31      |
| planar contexts           | 337.0    | 0.0       | 401_125 |
| iter_ruptures             | 166.6    | 0.15766   | 27_787  |
| ClassicalCalculator.run   | 96.4     | 1_210     | 1       |

# parallel tiling
| calc_69834, maxmem=7.7 GB | time_sec | memory_mb | counts  |
|---------------------------+----------+-----------+---------|
| total classical           | 1_230    | 74.3      | 31      |
| planar contexts           | 903.9    | 0.0       | 830_540 |
| iter_ruptures             | 532.4    | 0.14874   | 82_659  |
| ClassicalCalculator.run   | 179.9    | 1_234     | 1       |