Pool and JDC mishandle extranonce allocator exhaustion

[GitGab19]

While working on #729, I found out some similar issues on JDC and Pool, which I prefer to tackle with a specific PR.

Problem

Pool and JDC handle ExtranonceAllocator failures inconsistently:

• allocate_standard() failures trigger a global application shutdown.
• allocate_extended() reports every failure as min-extranonce-size-too-large, including CapacityExhausted.

This affects both applications, although JDC’s limit (65,536 channels) is lower than Pool’s (16,777,216).

Expected behavior

Distinguish InvalidRollableSize from CapacityExhausted, reject only the affected channel request with an appropriate OpenMiningChannelError, and keep existing connections running.

Add regression tests for both Pool and JDC.