Draw the rest of the ̶o̶w̶l̶ new ABI-handling framework

[workingjubilee]

brain recently tied together the rest of the connections necessary for implementing all of #1337 and thus finishing out the fixes that ultimately address the soundness issues revealed by the Spi bugs we found.

• BorrowDatum
  • better support zero-allocation datum conversion
  • prototype
  • identify unresolved design questions
• DetoastDatum
  • support datums with potentially very complex unboxing behavior
  • prototype
  • identify unresolved design questions
• MemCx
  • need significant API expansion
  • prototype
  • API expansion to make it actually usable
    • impl Allocator for MemCx #1737
    • need an answer for stable, pgrx-specific types
• Transition to new argument/return strategies that enable checking lifetimes
  • surface lifetimes into internal wrapper signatures so arguments/returns can use them
    • Manually staticize lifetimes in pgrx tests #1461
    • Demand lifetimes for pgrx::composite_type! #1462
    • Use composite_type! lifetimes in pgrx-sql-entity-graph #1463
    • Unbox types properly from arrays #1476
    • Allow real borrow-checking of #[pg_extern] fn lifetimes #1724
  • replacement for pg_getarg with a different signature for the new ABI, probably Implement VariadicAny around pg_sys::FunctionCallInfo #1428
  • try playing type-tetris with traits and blanket impls to implement returns via traits ( became Implement RetAbi and BoxRet #1701 )
  • try playing type-tetris with traits and blanket impls to implement argument-passing via traits ( became Make unboxing args sound with ArgAbi #1731 )
  • many unresolved design questions
• Retire FromDatum, once replacements are in place:
  • ArgAbi
  • BorrowDatum
  • DetoastDatum
• Spi: SpiConnect<'conn, 'ret: 'conn> #1483

[workingjubilee]

We still need to introduce some transition code here for cases where it isn't as obvious a breakage, but implementing the new RetAbi was more successful in preserving functionality than I expected, and most of the errors produced are straightforward type errors now, so I'm making more hard breaks.

[workingjubilee]

The first piece of this has slid into place, as #1731 now makes it possible, in conjunction with #1701, to have a single pg_extern function be lifetime-checked on its arguments and return types.