fix(#3491): support window union multiple join in request mode

[aceforeverd]

Always construct new physical nodes in a optimizer. The optimizer has a physical node cache builtin that prevent updating existing physical node in place.

When a physical node hit cache with the same node id, will break the optimize step afterwards: reconstructing schema context, which cause the issue in #3491

Ref

OpenMLDB/hybridse/src/passes/physical/group_and_sort_optimized.cc

Lines 440 to 448 in 0fa0fe9

	PhysicalOpNode* new_depend = nullptr;
	if (!KeysOptimized(request_join->GetProducer(0)->schemas_ctx(), request_join->GetProducer(0), left_key,
	index_key, right_key, sort, &new_depend)) {
	return false;
	}
	if (!ResetProducer(plan_ctx_, request_join, 0, new_depend)) {
	return false;
	}

[codecov]

Codecov Report

Patch coverage: 25.71% and project coverage change: +38.22% 🎉

Comparison is base (a013ba3) 37.14% compared to head (644215c) 75.37%.
Report is 2 commits behind head on main.

Additional details and impacted files

@@              Coverage Diff              @@
##               main    #3493       +/-   ##
=============================================
+ Coverage     37.14%   75.37%   +38.22%     
  Complexity      414      414               
=============================================
  Files           179      707      +528     
  Lines         10364   128663   +118299     
  Branches       1191     1191               
=============================================
+ Hits           3850    96976    +93126     
- Misses         6273    31446    +25173     
  Partials        241      241               

Files Changed	Coverage Δ
hybridse/include/vm/schemas_context.h	100.00% <ø> (ø)
hybridse/src/vm/runner.cc	69.25% <ø> (ø)
hybridse/src/vm/schemas_context.cc	85.85% <10.71%> (ø)
...se/src/passes/physical/group_and_sort_optimized.cc	77.44% <85.71%> (ø)

... and 524 files with indirect coverage changes

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