More C API compliance work

[DavisVaughan]

Quite a few R 4.5.0 helpers to make some more progress here. Commits are self contained.

Before / after

   File ‘rlang/libs/rlang.so’:
     Found non-API calls to R: ‘ENCLOS’, ‘OBJECT’, ‘PRENV’, ‘PRVALUE’,
       ‘R_PromiseExpr’, ‘Rf_allocSExp’, ‘Rf_findVarInFrame3’, ‘SETLENGTH’,
       ‘SET_BODY’, ‘SET_CLOENV’, ‘SET_ENCLOS’, ‘SET_FORMALS’,
       ‘SET_GROWABLE_BIT’, ‘SET_TRUELENGTH’, ‘SET_TYPEOF’, ‘XTRUELENGTH’


   File ‘rlang/libs/rlang.so’:
     Found non-API calls to R: ‘PRENV’, ‘PRVALUE’, ‘R_PromiseExpr’,
       ‘Rf_findVarInFrame3’, ‘SETLENGTH’, ‘SET_BODY’, ‘SET_CLOENV’,
       ‘SET_ENCLOS’, ‘SET_GROWABLE_BIT’, ‘SET_TRUELENGTH’

Fully dropped

• ENCLOS
• OBJECT
• Rf_allocSExp
• SET_FORMALS

Not shown but also dropped for vctrs via this

• BODY
• CLOENV

Remaining

• PRENV, PRVALUE, R_PromiseExpr (i.e. PREXPR) all need delayed expression utils from our patch
• Rf_findVarInFrame3 very complicated one tied up with the new R_GetVar and "missing" semantics
• SETLENGTH, SET_GROWABLE_BIT, SET_TRUELENGTH for RESIZE(), no known plan here yet
• SET_BODY in fn_zap_srcref(), could probably make a new function instead? We already do a clone of the function itself, so probably just as expensive. (Remove SET_BODY() usage #1845)
• SET_CLOENV tied up in on_exit_restore_lexical_env() but it looks like you can just make a new function here too? (Remove SET_CLOENV() usage #1846)
• SET_ENCLOS part of r_env_poke_parent() used in lots of tidy-eval things (Remove SET_ENCLOS() usage #1847)

May make follow up PRs for SET_BODY and SET_CLOENV, but they are more complicated so should be their own PRs

[DavisVaughan]

Here R_ClosureExpr() is the same as BODY_EXPR(), which I think only still exists because of us

https://github.com/wch/r-source/blob/4241940e50d0d60cd459cf162ddb5e8ecd0307b3/src/include/Rinternals.h#L1285

It's worth noting that R_ClosureBody() also exists which directly calls BODY(), but the previous BODY_EXPR() and R_ClosureExpr() also handle bytecode stuff by doing R_BytecodeExpr(R_ClosureBody(fn)), which I think we want, so I have not exposed the new R_ClosureBody() at all.