Add context_window override so the prompt %/compaction follow it live

[chauncygu]

The prompt's context-usage % and the compaction trigger derive from the model's context window, which previously could only be a hardcoded provider default — and max_tokens (the OUTPUT cap) does not change it. Add a dedicated, session-settable override.

• cc_config.py: new DEFAULTS key context_window (0 = use the model default).
• providers.context_window_override(config): single source that parses the override (positive int wins; 0/unset/negative/non-numeric/bool -> 0).
• compaction.get_context_limit(): honors the override, kept distinct from max_tokens. Bidirectional — smaller forces earlier compaction, larger corrects a stale default.
• providers.py send paths: apply the same override to the per-call output-cap window (dynamic_cap_max_tokens) so %, compaction, and the cap stay consistent. No-op when unset, so existing behavior is unchanged.
• /config context_window=: warns when it exceeds the model's real window (disables compaction -> API may reject oversized prompts) and notes it takes effect on the next prompt (no restart).

Read live each prompt, so changing model or context_window updates the % with no restart. Tests cover parsing, both-direction overrides, and that max_tokens is never read as the context window.