benchmark.sh: add missing commands for pre-release pipeline

[pszymkowiak]

Problem

scripts/benchmark.sh currently tests 22 commands but is missing ~12 testable commands that exist in RTK. This means regressions in those filters can slip through the pre-release pipeline undetected.

Missing commands (testable)

Command	Notes
rtk cargo build/test/clippy	Always available (we're a Rust project)
rtk smart <file>	2-line summary, easy to test
rtk diff <f1> <f2>	File diff, easy to test with temp files
rtk tree	Directory tree, easy to test
rtk curl <url>	HTTP fetch, test with public URL
rtk wget <url>	Download, test with public URL
rtk npm run	If package.json present
rtk npx	If package.json present
rtk proxy <cmd>	Passthrough, always available
rtk ruff check	Already has Python fixture, just uses wrong $RTK test wrapper
rtk pytest	Same — should use $RTK pytest not $RTK test pytest
rtk pip list	If pip available
rtk go test/build	Already has Go fixture, just uses wrong $RTK test wrapper
rtk golangci-lint	Same — should use $RTK golangci-lint not $RTK test golangci-lint
rtk format	Universal format checker

Not benchmarkable (skip)

gain, discover, learn, config, init, cc-economics, hook-audit — these are analytics/setup commands, not filters.

Current bugs in benchmark.sh

The Python and Go sections use $RTK test ruff check . and $RTK test golangci-lint run instead of the dedicated $RTK ruff check . and $RTK golangci-lint run commands. This means we're benchmarking the generic test runner instead of the specialized filters.

Expected outcome

All filter commands covered in benchmark.sh so we can catch token savings regressions before release.

# Pre-release validation
cargo build --release
bash scripts/benchmark.sh  # Should test ALL filters