[CI] add multithreaded matetrack runs that simulate gameplay

[robertnurnberg]

To be able to catch catch bugs like #6293 in future.

Notice that these runs are nondeterministic, like our other existing multithreaded matetrack runs. It is important to run these without TBs, since we cannot guarantee PV extension completion. The time settings were chosen so that the new tests take about the same time as the existing matetrack ones (around 2-3mins).

No functional change

[robertnurnberg]

Looks like the new CI fails for master? Unfortunate. That would mean #6636 did not fully fix the issue. Will try to reproduce locally.

Edit: CI failed on my fork, which was not rebased on latest master yet.

[robertnurnberg]

Found mate #-3 (better) for FEN "5rk1/ppp2p1p/4bB2/1nRr4/1P6/8/1P1Q1PPP/1R4K1 b - -" with bm #-7.
PV: f8b8 d2h6 d5d1 b1d1 e6d5 h6g7

Using /home/runner/work/Stockfish/Stockfish/Stockfish/src/stockfish on mates2000.epd with --time 3.0 --timeinc 0.01 --threads 4
Engine ID:     Stockfish dev-20260228-b4d45531
Total FENs:    2000
Found mates:   1376
Best mates:    966

Parsing the engine's full UCI output, the following issues were detected:
Better mates:        1   (from 1 FENs)

So definitely latest master. In fact, the CI ran on my fork. As evidenced by the bench of the binary used there:

Nodes searched  : 2396996