chore: add Codescythe stage profiling

[longlho]

Summary

• add opt-in profiling Cargo features for the core crate and CLI
• add //crates/codescythe_cli:codescythe_profiling as the Bazel profiling binary
• compile profiler timers, counters, and resolver hot-path branches out of the default binary
• keep CODESCYTHE_PROFILE=1 profile output on stderr for profiling builds
• document the current Kibana profiling findings and optimization priorities in ARCHITECTURE.md

Kibana Profile

Representative optimized run with the vendored Kibana fixture using //crates/codescythe_cli:codescythe_profiling:

• total: 12.18s
• project discovery: 1.18s
• reachable graph walk: 10.41s
• graph frontier parse: 4.03s
• graph frontier inspect: 6.38s
• issue map construction: 81ms
• JSON serialization: 12.9ms
• resolver calls: 1,715,145
• resolver cache hit rate: 75.9%
• uncached resolver time: 4.77s

The profiling run produced JSON output byte-for-byte identical to the baseline output.

Validation

• cargo fmt --check
• git diff --check
• bazel build -c opt //crates/codescythe_cli:codescythe
• bazel build -c opt //crates/codescythe_cli:codescythe_profiling
• binary string scan: default binary has no CODESCYTHE_PROFILE / profile output strings; profiling binary does
• cmp -s /private/tmp/kibana-baseline-output.json /private/tmp/kibana-stage-profile-feature-opt-output.json
• bazel test //crates/codescythe:codescythe_test //crates/codescythe_cli:codescythe_cli_test //crates/codescythe_cli:codescythe_cli_e2e_test //benchmarks:kibana_fixture_test