Parallel redesign

[Opt-Mucca]

This PR is a first redesign of the parallel MIP code. I'm just going to dot point the major changes:

• Adds nodes to the preparedNodes of each worker. A worker now only processes nodes assigned to it.
• Adds a max node processing limit of 100 to each worker. This tries to balance thread utilisation.
• Change backtrackPlunge to a simple DFS style algorithm without the ability to skip a level of the tree while backtracking based on pseudocosts.
• Change the nodequeue attached to each worker to a simple std::vector.

For my tests on 8 threads this gives a 6.5% time improvement and 6% LP iteration improvement. It gives a 1% performance degradation on single-threaded performance. Edit: This change should also use 40% less memory.

Reason for single-threaded degradation: There is one outlier instance tbfp-network where the sub-mip no longer finds the optimal solution instantly. This means the problems solves 5x slower. It stems from the backtrackPlunge changes, but I'd mostly chalk it up to randomness. Without this instance the change is performance neutral on single-threaded and more than 6.5% improving on 8 threads.

[codecov]

Codecov Report

❌ Patch coverage is 90.76087% with 17 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.12%. Comparing base (2ed99be) to head (ef56dde).
⚠️ Report is 277 commits behind head on latest.

Files with missing lines	Patch %	Lines
highs/mip/HighsMipSolver.cpp	92.12%	13 Missing ⚠️
highs/mip/HighsSearch.cpp	76.47%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           latest    #3192      +/-   ##
==========================================
+ Coverage   73.06%   73.12%   +0.06%     
==========================================
  Files         445      445              
  Lines      107661   107801     +140     
  Branches    17257    17262       +5     
==========================================
+ Hits        78659    78826     +167     
+ Misses      28726    28699      -27     
  Partials      276      276              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[fwesselm]

@Opt-Mucca, this looks great! Just a couple of uninformed comments from me.

[fwesselm]

The return value of evaluateNode is ignored here? It could return true, right? Does this overcount the total_nodes_explored?

[Opt-Mucca]

I don't think this overcounts total_nodes_explored. The node should be processed by pruneNode and then increment the counter. Edit: The node is already sent of to processedNodes, but pruneNode should recognise this.

The bool return seems to be something left-over from early drafts (isn't even used on master currently)