Summary
Changing transaction format or ID would create a precedent which would affect all non-node software that processes raw transactions. Because it is unknowable what software relies on the assumption of a fixed transaction format and ID, breaking that assumption would have costs and risks that are difficult to assess and contain where only the lower bound could be established by gathering evidence.
Transaction format can be upgraded in a way which would have predictable costs and risks. Most of the known desired features from transaction format changes can be introduced in a way with better contained costs and risks.
Because there exists a knowable node-contained-cost alternative, there is a natural deterrent to changing the format: it would be a greater burden on the one who pushes for a change than it would be to adapt his change to the alternative approach.