fix(bench): improve benchmark stability and output formatting

[iduartgomez]

Fixes three issues with benchmark reports:

1. ANSI escape codes in output - Cargo warnings with color codes were
   polluting markdown reports. Fixed by:

   • Adding strip_ansi_codes() to parse_bench_output.py
   • Using --color=never in both workflow files
   • Adding sed fallback stripping in workflow error handling

2. Benchmark panic during warmup - ConnectionClosed errors in warmup
   phase caused benchmark crashes. Fixed by:

   • Replacing .unwrap() with graceful error handling in slow_start.rs
   • Making ConnectedPeerPair::warmup() return completion count instead
     of panicking on errors

3. Extended benchmarks not running - transport_extended was missing from
   Cargo.toml [[bench]] configuration, causing cargo to run test harness
   instead of criterion. Fixed by adding proper bench configuration with
   harness = false.