Allocation manager inference depends on the classpath location name #1469

Describe the bug, including details regarding any error messages, version, and platform.

When no allocation manager type is set, CheckAllocator.check() finds a DefaultAllocationManagerFactory resource on the classpath, then picks the factory class by searching the resource location for memory-core, memory-unsafe or memory-netty, in that order.

This has a few issues:

• The location name decides the factory, not the resource that was actually found. A Netty or Unsafe factory loaded from a location that contains memory-core (a directory or a repackaged jar name) is inferred as the memory-core test factory, and creating a RootAllocator fails with Unable to instantiate Allocation Manager for org.apache.arrow.memory.DefaultAllocationManagerFactory.
• When both arrow-memory-unsafe and arrow-memory-netty are on the classpath, Unsafe is chosen without any warning.
• The log prints the factory class name, not the Netty / Unsafe value accepted by arrow.allocation.manager.type / ARROW_ALLOCATION_MANAGER_TYPE, and the inference is logged on two lines (CheckAllocator and DefaultAllocationManagerOption).
• The location in the CheckAllocator log is cut at the first memory found in the path.

I expect the factory to be inferred from the resource that was found, a warning when several allocation manager types are on the classpath, and a single log line with the inferred type and its location.

Follow-up to #1461.