Relax template-haskell upper bound for persistent

[mpilgrem]

GHC 9.6.1 ships with template-haskell-2.20.0.0.

Tested by building the persistent library on Windows with Stack and the stack.yaml:

resolver: lts-20.15

packages:
  - ./persistent
  - ./persistent-sqlite
  - ./persistent-test
  - ./persistent-mongoDB
  - ./persistent-mysql
  - ./persistent-postgresql
  - ./persistent-redis
  - ./persistent-qq

extra-deps:
# template-haskell-2.20.0.0 is currently only a candidate package on Hackage
- url: https://hackage.haskell.org/package/template-haskell-2.20.0.0/candidate/template-haskell-2.20.0.0.tar.gz
- ghc-boot-th-9.6.1
# Accepts template-haskell-2.20.0.0
- aeson-2.1.2.1
- generically-0.1.1
- tagged-0.8.7
# GHC boot packages
- bytestring-0.11.4.0
- text-2.0.2
- binary-0.8.9.0
- exceptions-0.10.7

[parsonsmatt]

Do you mind adding GHC 9.6 to the CI for this? Code here: #1485

Tempted to merge as-is but would like to know it's safe

[mpilgrem]

@parsonsmatt, I've updated the CI matrix. I took the liberty of also bumping the GHC 9.2.x and 9.4.x series. I assumed, perhaps wrongly, that GHC 9.6.1 and Cabal 3.10.1.0 go hand in hand.

[mpilgrem]

I see you have already taken action on the cryptohash-md5 dependency and GHC 9.6.1: haskell-hvr/cryptohash-md5#15

[mpilgrem]

There is a revision 2 to cryptohash-md5-0.11.101.0 on Hackage, but for some reason the CI did not appear to acknowledge it.