faster SRCINFO parsing + memory-based building

Switched to parsing srcinfo with channels on all available cpus, speeding
up srcinfo-parsing and queue generation by a lot.

New memory-limit based building will max out the available memory while
not building the same packages at the same time for different marchs,
fixing some long-standing bugs like firefox not building at the same time
because the same ports are used for profile-based optimization. This
also drops the artificial delay on build-start, speeding up things even
more. This also means there is no hard-coded limit on how many packages
can be build at once anymore. As long as there is RAM available, builds will be
started.

ent/dbpackage/where.go

@@ -6,7 +6,7 @@ import (
 	"time"
 
 	"entgo.io/ent/dialect/sql"
-	"git.harting.dev/ALHP/ALHP.GO/ent/predicate"
+	"somegit.dev/ALHP/ALHP.GO/ent/predicate"
 )
 
 // ID filters vertices based on their ID field.