gh-112287: Speed up Tier 2 (uop) interpreter a little (#112286)

This makes the Tier 2 interpreter a little faster.
I calculated by about 3%,
though I hesitate to claim an exact number.

This starts by doubling the trace size limit (to 512),
making it more likely that loops fit in a trace.

The rest of the approach is to only load
`oparg` and `operand` in cases that use them.
The code generator know when these are used.

For `oparg`, it will conditionally emit
```
oparg = CURRENT_OPARG();
```
at the top of the case block.
(The `oparg` variable may be referenced multiple times
by the instructions code block, so it must be in a variable.)

For `operand`, it will use `CURRENT_OPERAND()` directly
instead of referencing the `operand` variable,
which no longer exists.
(There is only one place where this will be used.)