One tooling language: the first checker moved to Rust

Five languages, and only one of them removable. C is the product and freestanding by mandate; GLSL is forced by the GPU probes; sh is glue and a documented choice over PowerShell; Rust is already paid for at 8,194 lines, 17 subcommands and 133 tests. Python's 4,358 lines are the only ones whose job another language in this repo already does.

Python is not a build dependency - the Makefile never invokes it - so README.md's "the only build dependency is clang" is honest. It is a verification dependency, which is worse: the gates were the only part of this tree written without tests or types.

scripts/guards.py is now obscene-tool guards, and the port paid for itself before it was finished. Run side by side against the same tree the two disagreed: 128 checks against 135. Neither was right.

• The Python required the symbol column to match "(\w+)", so it silently skipped eight real rows - every check whose symbol is descriptive rather than an identifier: "(self-check)", "mapped memory", "(census control)", "(compute dispatch)", "(every censused symbol)". Among them 910-bulk/probe, the blind prober, which calls arbitrary censused symbols and is the check a guard rule least wants to miss.
• The first Rust version over-matched in the other direction, counting obs_report_progress("910-bulk/probe", ...) as a table row because it looks like one.
• Both assumed every runner is named check_*. The blind prober's is run_bulk. The runner is a positional field; a naming convention is a habit, not a contract.

Anchoring on OBS_CAP - the token that actually makes a row a row - gives 136 of 136, matching an independent count. Nine tests now pin each of those failures, including that a symbol named in a comment is not a call, and that division is not the start of one.

Neither Python fault could ever have appeared as a failure. Both appeared as a smaller number nobody was comparing against anything, which is this project's most-repeated defect and the whole argument for the move.

And scripts/lint.sh had the . bug. . "$HOME/.cargo/env" 2>/dev/null || true - a POSIX special built-in that fails exits the shell before || is consulted, so on a machine without that file the script exited with no output and status 1, indistinguishable from a tree full of lints. verify.sh was fixed for exactly this months ago; this file was never run on the host, so it kept the bug. Found by running it on the host.

Sixteen Python scripts remain. The order is checkers, then analysis, then generators.