Add support MasakhaNER v2 dataset

[stefan-it]

Hi,

this PR adds support for the recently released version 2 of the MasakhaNER dataset!

A new version argument was added and now defaults to "v2" of the dataset. Demo usage:

from flair.datasets import NER_MASAKHANE

v1_all = NER_MASAKHANE(languages="all", version="v1")
v2_all = NER_MASAKHANE(languages="all", version="v2")

Closes #2971.

Unittests

This PR also adds some unittests for v1 and v2 of the dataset. Here's a comparison between the number of sentences, that are mentioned in the paper and the "Parsed" ones with this implementation:

Version 1 - bold denotes difference:

Language	Paper (Train, Dev, Test)	Flair (Train, Dev, Test)
amh	1750 / 250 / 500	1750 / 250 / 500
hau	1903 / 272 / 545	1903 / 272 / 545
ibo	2233 / 319 / 638	2233 / 319 / 639
kin	2110 / 301 / 604	2110 / 301 / 604
lug	2003 / 200 / 401	2003 / 200 / 401
luo	644 / 92 / 185	644 / 92 / 185
pcm	2100 / 300 / 600	2100 / 300 / 600
swa	2104 / 300 / 602	2104 / 300 / 602
wol	1871 / 267 / 536	1871 / 267 / 536
yor	2124 / 303 / 608	2124 / 303 / 608

Version 2 - bold denotes difference:

Language	Paper (Train, Dev, Test)	Flair (Train, Dev, Test)
bam	4462 / 638 / 1274	4462 / 638 / 1274
bbj	3384 / 483 / 966	3384 / 483 / 966
ewe	3505 / 501 / 1001	3505 / 501 / 1001
fon	4343 / 621 / 1240	4343 / 621 / 1240
hau	5716 / 816 / 1633	5716 / 816 / 1633
ibo	7634 / 1090 / 2181	7634 / 1090 / 2181
kin	7825 / 1118 / 2235	7825 / 1118 / 2235
lug	4942 / 706 / 1412	4942 / 706 / 1412
mos	4532 / 648 / 1294	4532 / 648 / 1294
pcm	5646 / 806 / 1613	5646 / 806 / 1613
nya	6250 / 893 / 1785	6250 / 893 / 1785
sna	6207 / 887 / 1773	6207 / 887 / 1773
swa	6593 / 942 / 1883	6593 / 942 / 1883
tsn	3489 / 499 / 996	3489 / 499 / 996
twi	4240 / 605 / 1211	4240 / 605 / 1211
wol	4593 / 656 / 1312	4593 / 656 / 1312
xho	5718 / 817 / 1633	5718 / 817 / 1633
yor	6877 / 983 / 1964	6876 / 983 / 1964
zul	5848 / 836 / 1670	5848 / 836 / 1670

[alanakbik]

@stefan-it thanks for adding this!