Make top-p configurable via .sgptrc (closes #425)

[gandzekas]

PR: Make top-p (top-probability) configurable via .sgptrc — closes #425

Summary

Issue #425 asks to set a default temperature and top-probability via the
runtime config file (~/.config/shell_gpt/.sgptrc).

• Temperature config support already landed in Added default temperature to config #723 (DEFAULT_TEMPERATURE).
• This PR completes the issue by making top-p config-driven too, via a new
  DEFAULT_TOP_P key. Previously top_p was hardcoded to 1.0 in app.py, so
  the config file had no effect on it.

Changes

1. sgpt/config.py — add DEFAULT_TOP_P to DEFAULT_CONFIG (default 1.0,
   overridable via DEFAULT_TOP_P env var, consistent with DEFAULT_TEMPERATURE).
2. sgpt/app.py — top_p typer.Option default now reads cfg.get("DEFAULT_TOP_P")
   instead of the literal 1.0.
3. README.md — document both DEFAULT_TEMPERATURE and DEFAULT_TOP_P in the
   runtime config example block (temperature was previously undocumented).

Behavior

• Default behavior is unchanged: DEFAULT_TOP_P defaults to 1.0, identical to
  the previous hardcoded value.
• Users can now set DEFAULT_TOP_P=0.9 (or any 0.0–1.0 value) in .sgptrc to
  apply a default nucleus-sampling probability without passing --top-p every run.
• Explicit --top-p on the command line still overrides the config, as before.

Verification

• python3 -m py_compile sgpt/config.py sgpt/app.py passes.
• Existing test suite passes (run in isolated venv:
  OPENAI_API_KEY=dummy sgpt-venv/bin/python -m pytest tests/ -q).

Test plan for maintainers

• sgpt "hello" with no --top-p uses DEFAULT_TOP_P from config (default 1.0).
• Setting DEFAULT_TOP_P=0.5 in .sgptrc changes the default without flags.
• --top-p 0.3 still overrides the config value.