fix(core): avoid 0% context after /compact

[ozan-s]

Fixes #9416.

Repro

1. Run /compact in the TUI.
2. Immediately after the “Context compacted” message, the footer shows:
   0% context left · ? for shortcuts

Expected

After compaction frees space, the footer should show a non-zero context percentage remaining.

Root cause

After remote compaction, the history includes a compaction_summary (ResponseItem::Compaction) with encrypted_content that can be very large. Core recomputes token usage post-compaction via Session::recompute_token_usage, which calls ContextManager::estimate_token_count. The estimator was treating the full encrypted_content length as prompt-sized text, producing an estimated total >= context window; the UI then clamps to 0% remaining.

Fix

Cap the token estimate for ResponseItem::Compaction.encrypted_content to a small, context-window-relative bound (10% of the effective window, capped at 25k tokens). This keeps post-compaction estimates realistic (and avoids transient 0%), without special-casing the UI.

Tests

Add a regression unit test that constructs a compacted history with a huge encrypted_content blob and asserts the estimated total stays below the configured context window.

[etraut-openai]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".