fix(gateway): bound lifecycle scans of binary paths

[poctek]

Summary

• distinguish missing referenced scripts from local binaries before remote fallback
• skip NUL-bearing local and remote binary content instead of recursively tokenizing it
• handle NUL-bearing candidate paths without raising ValueError or calling the remote reader
• cap remote script reads at 1 MiB + 1 byte and fail closed on oversized text
• preserve remote scanning for scripts that are genuinely absent on the host

Why this is still needed after #77703

#77703 prevents the eventual ValueError and skips NUL-bearing output, but the remote fallback still uses an unbounded cat. For a local executable larger than 1 MiB, _read_script_in_env() skips its bounded local branch and then cats the same file through the environment. The NUL check only runs after the entire binary has been captured.

This PR fixes that remaining memory/hang path by distinguishing missing from present but binary, and by bounding genuinely remote reads. In production, a 273 MB Rust debug ELF drove the gateway to multi-GB memory use and stalled it before the executable ran.

Root cause

_read_referenced_script() returned None both when a local path was missing and when it had positively identified a NUL-bearing binary. The caller interpreted both states as remotely missing and invoked read_remote_script. Path-based executables could therefore be read again and fed into the lifecycle regex/path walk.

The tri-state result now permits remote fallback only when os.open() could not open the path. Once a local path has opened, binary detection or a later read failure cannot reclassify it as remotely missing.

Tests

• uv run --extra dev pytest -q tests/hermes_cli/test_gateway_restart_loop.py tests/tools/test_terminal_tool.py tests/cron — 493 passed
• uv run --extra dev ruff check cron/lifecycle_guard.py tools/terminal_tool.py tests/hermes_cli/test_gateway_restart_loop.py tests/tools/test_terminal_tool.py — passed
• python -m py_compile for all modified Python files — passed
• real 273 MB target/debug/solana-deploy lifecycle-guard harness — allowed in 0.004s with no remote callback

One pre-existing cron test warning remains: _send_to_platform coroutine was never awaited.

Fixes #78800
Related: #76762, #77703, #78942

[kshitijk4poor]

Closing as superseded by #80258, which fixes this whole bug class architecturally rather than per-callsite: path candidates are sanitized once at the ingestion boundary (NUL/empty/unexpandable tokens rejected before any OS call), text from any read_remote_script callback is sanitized at the recursion boundary (NUL = binary = nothing to scan; >1 MiB = fail closed), the remote fallback read is bounded at the source (head -c, so oversized binaries never cross the wire), and the public guard is total by construction — an unexpected walk failure logs and falls back to the direct-scan verdict instead of breaking every terminal command.

Your report and fix targeted a real member of this class — thank you. The per-callsite patches kept leaving sibling frames exposed (#76762 → #77703 → #77780 → #78256 each crashed one frame away from the previous fix), which is why we went with the boundary fix instead of merging the fragments individually. #80258 carries regression tests for the NUL-path, binary-callback, oversized-read, unset-HOME, and walk-crash cases plus an adversarial never-raises sweep.