Implement BGP paths limit capability

[jamestiotio]

Implement the capability as specified by the draft-abraitis-idr-addpath-paths-limit-04 document.

Corresponding unit tests are also added to validate the functionality of this feature.

[jamestiotio]

This closes #1218. First time contributing to ExaBGP, so apologies for any beginner mistakes.

[thomas-mangin]

Thank you @jamestiotio
I will look into it in the next few days.

[thomas-mangin]

The code was merged but then changed to match a different configuration format (and some issues were found by Claude which resolved them).

[thomas-mangin]

Only for information, the findings were:

🔴 prefix_index() only implemented for INET

NLRI.prefix_index() falls back to self.index(), which includes the path-ID. For VPN, EVPN, FlowSpec, and other AddPath-capable families, paths-limit enforcement and audit will silently malfunction (each path-ID treated as a separate prefix, so the limit is never hit). Either implement prefix_index() on other NLRI subclasses or document this as an IPv4/IPv6 unicast-only feature.

🟡 Unbounded memory in _path_counts/_path_warned

IncomingRIB._path_counts grows with every unique prefix seen over the session lifetime. Withdrawn prefixes are cleaned up, but a peer that announces-then-withdraws millions of distinct prefixes will accumulate warned-set entries until clear(). Not critical since audit is optional, but worth noting.

🟡 unpack_capability loop repeats data = data[5:] three times

pathslimit.py:250,256,259 - could restructure to slice once at loop end. Minor readability nit.

🟡 Configuration output in neighbor.py:351-354

The paths-limit block emission mixes + concatenation with f-string implicit concatenation. Works, but fragile if someone later edits the surrounding f-string.

[thomas-mangin]

Thank you for your help.