perf: avoid repeated export line scans

[longlho]

Summary

• avoid repeatedly scanning from the start of a source file for every exported symbol's line/column
• keep the existing single-export path unchanged, and build a per-file line-start table only for files with multiple exports
• add a parser unit test that checks the line-start lookup against the original scanner, including non-ASCII text

Benchmark Notes

Kibana fixture, --samples 3 --warmups 1 --skip-knip --skip-build:

Build	Codescythe
origin/main baseline	13988.2ms +/-10.57%
this branch	13401.0ms +/-2.55%

The optimized Kibana JSON output was byte-for-byte identical to the origin/main output for the same config.

Validation

• git diff --check
• bazel build -c opt //crates/codescythe_cli:codescythe
• cmp -s /private/tmp/kibana-baseline-output.json /private/tmp/kibana-linecol-output.json
• bazel test //crates/codescythe:codescythe_test //crates/codescythe_cli:codescythe_cli_test //benchmarks:kibana_fixture_test