protocol: PayGas factorizes gas updates for instrs with static costs

In a Michelson script, most of the instructions have a statically determined cost in terms of gas consumption. We can update the gas counter and check for its exhaustion at the beginning of any sequence of such instructions:

• this is safe because we may only stop earlier than previously in case of gas exhaustion, and
• this is also better for performance. Indeed, preliminary benchmarks show that the Michelson interpreter spends a lot of time updating and checking the gas level. If we reduce the frequency of gas-related operations, we have more CPU time to actually execute the script instructions.

We proceed through the following steps:

1. The introduction of PayGas (n, i), a new Michelson instruction to pay n milligas before executing i without monitoring gas. (Changes on Script_typed_ir.)
2. The introduction of a preprocessing phase that inserts PayGas on appropriate sequences of instructions.
3. The introduction of a gas_monitoring_status in the Michelson step function.

We had to separate the cost_of_instr function in two functions static_cost_of_instr and dynamic_cost_of_instr. To avoid any code duplication between these two functions, we introduce a new parameter to instr which statically tells if the cost of an instruction can be statically determined.

This optimization is part of a sequence to optimize gas update and check for gas exhaustion documented here: https://hackmd.io/@yrg/rkLwB17wD