test: add Concore class API tests for concore.hpp

[avinxshKD]

Added TestConcoreHpp.cpp following the same custom pass/fail pattern. Sets up temp port dirs, does real file read/write round-trips, tests unchanged() state transitions, retry exhaustion on empty files, and param loading. No gtest.

closes #484

[avinxshKD]

Hey @pradeeban @Rahuljagwani @mayureshkothare

While building out the transport layer in the julia prototype, I ran into a performance question worth flagging

With file I/O, even delay=0 still hits disk on every poll (open + read + close per call). For co-located nodes, Mmap.jl would keep reads in memory -- same polling loop, no syscall overhead, so the delay can be pushed much lower without thrashing I/O.

Planning to measure this properly: file vs mmap vs ZMQ, same study, same workload. The benchmark results and methodology could double as a short research note if that's useful for the project. Any suggestions :p

Also should the mmap backend be opt-in at init, or would automatic detection of co-located nodes be worth the complexity?

[pradeeban]

yay, some research outcome is always nice. A lot of our GSoC contributions end up as research papers, including the GSoC 2025 project on concore.

I will vote for the performance over complexity in this case. If your added complexity gives you a decent performance boost, go for it.