Option to delete equal successive tempo / time sig changes

[Natooz]

When preprocessing a MIDI files, the previous version automatically deleted the successive tempo changes and time signature changes that had the same value.
Although this can make sense and reduce the nb of tokens, it can also reduce the recurrence of this information in sequences of tokens, where an user could on the other hand want to repeat it at recurrent intervals, especially with tokenizations such as TSD or MIDILike.

This option allows to decide wether or not to delete these identical successive changes.

[Natooz]

Still changes to do towards Octuple and CPWord, in order to robustly test their tempo and time signature encoding / decoding.

[Natooz]

@ilya16

After analyzing 5k MMD files + the Maestro + POP909 datasets, here is the count of Time signatures :

(3, 1): 1 (0.01 %)
(20, 32): 1 (0.01 %)
(10, 32): 1 (0.01 %)
(17, 16): 1 (0.01 %)
(18, 16): 1 (0.01 %)
(14, 16): 1 (0.01 %)
(3, 16): 1 (0.01 %)
(2, 16): 1 (0.01 %)
(6, 2): 1 (0.01 %)
(1, 2): 1 (0.01 %)
(7, 16): 1 (0.01 %)
(17, 8): 1 (0.01 %)
(1, 1): 1 (0.01 %)
(16, 8): 1 (0.01 %)
(14, 8): 1 (0.01 %)
(12, 16): 2 (0.02 %)
(16, 16): 2 (0.02 %)
(9, 4): 2 (0.02 %)
(16, 4): 2 (0.02 %)
(13, 8): 2 (0.02 %)
(4, 1): 2 (0.02 %)
(1, 16): 2 (0.02 %)
(14, 4): 2 (0.02 %)
(9, 16): 3 (0.04 %)
(11, 8): 3 (0.04 %)
(15, 8): 3 (0.04 %)
(8, 8): 4 (0.05 %)
(1, 8): 5 (0.06 %)
(4, 2): 6 (0.07 %)
(3, 2): 6 (0.07 %)
(11, 16): 6 (0.07 %)
(10, 8): 6 (0.07 %)
(4, 8): 8 (0.10 %)
(2, 8): 8 (0.10 %)
(7, 4): 11 (0.14 %)
(2, 2): 14 (0.17 %)
(5, 8): 14 (0.17 %)
(8, 4): 14 (0.17 %)
(7, 8): 21 (0.26 %)
(9, 8): 24 (0.30 %)
(3, 8): 52 (0.65 %)
(12, 8): 56 (0.70 %)
(5, 4): 65 (0.81 %)
(6, 4): 80 (0.99 %)
(6, 8): 126 (1.56 %)
(3, 4): 209 (2.59 %)
(2, 4): 521 (6.47 %)
(1, 4): 743 (9.22 %)
(4, 4): 6018 (74.69 %)
Time signature above 0.5%:
[(3, 8), (12, 8), (5, 4), (6, 4), (6, 8), (3, 4), (2, 4), (1, 4), (4, 4)]
{8: [3, 12, 6], 4: [5, 6, 3, 2, 1, 4]}

At the end, I printed the TS which represent at least 0.5% of all TSs. I resorted to keep these values for now. Should we add some more ?

[ilya16]

These seem to be the most common ones. I can also refer to the time signature analysis of Lakh MIDI Dataset.

Maybe adding 2/2, 9/8, 5/8, and 7/8 will make a nearly complete list.