READS_FROM/WRITES_TO/FLOWS_TO: extend I/O and data-flow edges beyond Python

[vitali87]

Summary

READS_FROM / WRITES_TO / FLOWS_TO are emitted for Python only. Every other indexed language parses into the graph with full structure/calls but produces zero I/O or data-flow edges, because the source/sink/handle registry and the processors' language gate are Python-specific.

Root cause (grounded)

• Both processors early-return for non-Python:
  • IOAccessProcessor.process_io_for_caller — if language != cs.SupportedLanguage.PYTHON: return (codebase_rag/parsers/io_access/processor.py:64).
  • FlowProcessor.process_flow_for_caller — same gate (codebase_rag/parsers/flow_access/processor.py:116).
• The registries have only Python entries: IO_SINKS, IO_HANDLE_CONSTRUCTORS, IO_HANDLE_METHODS are keyed {SupportedLanguage.PYTHON: (...)} (codebase_rag/parsers/io_access/registry.py).
• The AST-node handling (assignment / call / return / handle-method detection, e.g. _binding_from_node, _arg_names, _return_value_nodes) is written against Python tree-sitter node types (TS_PY_*).

Scope

Add per-language source/sink/handle tables plus per-language node-type handling. Rough per-language surface:

• Go — os.Getenv/os.Open/os.Create, net/http (http.Get, http.Post), database/sql (db.Query, db.Exec), fmt.Println.
• JavaScript / TypeScript — fs.readFile/writeFile, fetch/axios, process.env, console.log.
• Java — java.io/java.nio streams, java.sql (Statement.execute*), System.getenv, System.out.print*.
• Rust — std::fs, std::net, std::env::var, println!.
• C++ — fstream, sockets, getenv, std::cout.

Each language also needs its own node-type constants for assignment/call/return and handle-method receiver parsing, mirroring the Python paths.

Suggested direction

• Generalize the registry maps to {SupportedLanguage: (...)} (the dict shape already exists) and populate per language.
• Abstract the Python-specific node-type lookups behind a small per-language descriptor (assignment/left/right, call/function/arguments, return value nodes, attribute/handle receiver), so the processors' walk is language-agnostic.
• Land one language at a time behind the existing opt-in io capture group, each with its own edge tests. Direction heuristics (e.g. the SQL first-keyword refinement) should be revisited per language.

Until then, non-Python I/O/flow is silently absent; this issue tracks closing that gap.