Add configurable shell-history suppression

[tony]

Summary

• Add LIBTMUX_SUPPRESS_HISTORY=0|1 as a startup default for run_command only, with invalid values failing startup without reflecting configuration contents; keep suppress_persistent_history on create_session, create_window, split_window, and respawn_pane explicit per call and independent of that setting.
• Publish the effective non-nullable suppress_history default for run_command through tools/list; explicit per-call values win, and direct Python calls remain default-off.
• Apply best-effort history controls to spawned shells while preserving raw terminal input, paste behavior, and direct tmux launch strings.
• Document shell-specific limits, spawn scope, audit behavior, and the remaining visibility surfaces without presenting history suppression as secret transport.

Changes

MCP policy and contract

• Resolve the startup setting once for run_command and apply precedence as explicit call value, startup setting, then False; spawn tools never inherit it.
• Use FastMCP's public tool transform so eligible MCP schemas expose the effective default without changing direct Python signatures.
• Report the active policy in bounded server instructions while keeping raw send, batch, and paste behavior explicit.

Spawned shells

• Copy and normalize caller environment input before adding an empty HISTFILE, merging Bash's HISTCONTROL, and enabling Fish private-history variables.
• Reject conflicting history controls with value-free errors and never retry a failed call without suppression.
• Store create_session controls in the tmux session environment for future panes; keep create_window, split_window, and respawn_pane controls scoped to the spawned process.
• Leave direct shell and window_shell launch strings unchanged. An explicit false adds no controls but cannot remove controls already inherited from a session, parent process, or startup file.

Terminal safety

• Prefix the grouped run_command event only when suppression is effective and reject caller-supplied carriage returns or line feeds before tmux receives input; explicit false preserves intentional multiline commands.
• Keep send_keys and every send_keys_batch operation explicit-only so control keys, partial input, literal text, Enter boundaries, and TUI interaction retain their exact bytes.
• Keep paste_text and paste_buffer outside the suppression policy.

Documentation and audit behavior

• Describe Bash, Zsh, and Fish persistence and in-memory limits with version-pinned public sources.
• Distinguish mapping-form environment audit summaries from JSON-string scalar redaction.
• State that pane output, scrollback, capture, process inspection, client transcripts, hooks, and other logs remain observable.

Design decisions

• Use startup configuration plus per-call overrides instead of a mutable global tool, avoiding cross-client policy races.
• Apply the global default only to semantic run_command calls; spawned-shell controls stay explicit per call, and arbitrary terminal input cannot safely inherit a byte prefix.
• Treat suppression as best effort because shell startup files and user hooks can override history behavior after spawn.

Test plan

• Cover startup parsing, schema defaults, explicit overrides, direct and generic batch calls, and instruction budgeting.
• Prove raw control keys, literal and TUI input, timed and untimed batches, Enter boundaries, and paste paths remain unchanged.
• Exercise adversarial multiline rejection before tmux access and preserve explicit suppression-off multiline behavior.
• Cover environment copying, normalization, merging, conflict confidentiality, spawn scope, direct launch strings, and inherited-control behavior.
• Verify Bash, Zsh, and Fish mechanism application and persistence behavior with explicit skip reasons when a shell is unavailable.
• Contract-test configuration, tool pages, safety guidance, audit shapes, and the unreleased changelog.

Verification

• Ruff formatting and linting
• Strict mypy
• Full pytest suite
• Sphinx documentation build
• Production FastMCP inspection with suppression disabled and enabled
• Real FastMCP client and isolated tmux smoke
• Supported tmux matrix from 3.2a through master

Closes #90.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 95.71429% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.20%. Comparing base (e9fa03d) to head (0c83e18).

Files with missing lines	Patch %	Lines
src/libtmux_mcp/server.py	84.21%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #91      +/-   ##
==========================================
+ Coverage   85.00%   85.20%   +0.19%     
==========================================
  Files          43       44       +1     
  Lines        3202     3265      +63     
  Branches      439      452      +13     
==========================================
+ Hits         2722     2782      +60     
- Misses        349      352       +3     
  Partials      131      131              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.