Use proper TOML parser for Cargo workspace parsing

[lambda-alpha-labs]

Problem

The parse_cargo_toml function in src/extract/ci.rs uses manual line-by-line string parsing to extract test targets and workspace structure. While stable for typical configurations, this approach can fail on:

• Multiline arrays (e.g., long workspace member lists)
• Inline tables
• Complex nested TOML structures common in enterprise monorepos

Current code

else if trimmed.starts_with("[[test]]") {
    in_test_section = true;
    has_test_targets = true;
}

Proposal

Add the toml crate as a dependency and fall back to proper TOML deserialization when the line-based parser cannot extract a valid structure. The line-based parser should remain as the fast path for typical configurations.

Acceptance

• Complex TOML workspace structures parse correctly
• No regression on simple Cargo.toml files (existing tests pass)
• Fast path (line-based) remains default for trivial cases