Speed-up JSON infra by filling a subset of test and executing it

[marioevz]

Description

Reduce the time JSON Infra CI takes and also improve its accuracy, in particular for eips/ branches:

• Mark a small subset of tests (e.g., “pytest.mark.json_infra”) to cover each transaction type, format, and fork
• In the JSON infra job, first run fill ... -m "json_infra" ... to fill only those marked tests
• Immediately feed that freshly generated output into the JSON infra runner

Pre-requisites

#1537