fix(ci): improve test stability on high-core-count runners

[sanity]

Problem

CI tests have been flaky on self-hosted runners with high core counts (64 CPUs on technic). Multiple different tests fail intermittently with timeouts:

• test_put_contract
• test_put_contract_three_hop_returns_response
• test_put_merge_persists_state
• test_update_contract
• test_three_node_network_connectivity
• test_logger_capture*

Additionally, the six-peer-regression job fails with a linker error when building freenet fresh:

rust-lld: error: unable to find library -llzma

Root Cause

1. Test thread contention: Cargo defaults to one test thread per CPU. With 64 CPUs, 64+ network tests run simultaneously, causing:

   • UDP port exhaustion/collisions
   • Async runtime starvation
   • Timing-sensitive operations failing their timeouts

2. Library path not set: When six-peer-regression builds freenet from river's workspace (fresh build, no cache), the linker can't find liblzma because rust-lld doesn't search /usr/lib by default on Arch Linux.

Solution

1. Limit test threads to 8 (--test-threads=8) - reduces parallel test execution to a manageable level
2. Set LIBRARY_PATH and PKG_CONFIG_PATH for the six-peer-regression job

Testing

• All tests pass consistently with thread limiting (verified on both nova and technic)
• Currently running CI on the hop-by-hop-routing PR with these same fixes

[AI-assisted - Claude]