internal/tests: Fix double generate

[abhinav]

The tests in internal/tests currently have a hacky "invoke cff twice" setup:

• once to generate the code for the test that needs --auto-instrument
• once to generate everything else

The test that relies on auto instrument is tagged with autoinstrument,
but we generate everything with auto-instrument,
and then replace it in the second call.

This is inefficient and we should fix it.