ci: constrain megaparsec < 9.8 on GHC < 9.10

[jkeuhlen]

While trying to fix a CI issue in another PR (which came up on the actual functionality PR I was working on), I ran into this update to megaparsec which causes it to fail to build on older GHC versions. This pins the constraint for older versions so that everything can keep building.

Claude summary below, in case it's useful.

9.8 also causes test failures due to some other changes, so rather than a conditional fix I just pinned it below 9.8 everywhere:

  test/Database/Persist/QuasiSpec.hs:1261:17: 
  1) Database.Persist.Quasi, tab error level setting, when configured to disallow tabs, rejects tab indentation
       expected: 2:1:
                   |
                 2 |  Id Text
                   | ^
                 unexpected tab
                 expecting valid whitespace
                 
                 3:1:
                   |
                 3 |  name String
                   | ^
                 unexpected tab
                 expecting valid whitespace
                 
        but got: 2:1:
                   |
                 2 |  Id Text
                   | 
                 unexpected tab
                 expecting valid whitespace
                 
                 3:1:
                   |
                 3 |  name String
                   | 
                 unexpected tab
                 expecting valid whitespace
                 

  To rerun use: --match "/Database/Persist/Quasi/tab error level setting/when configured to disallow tabs/rejects tab indentation/" --seed 608015465

Problem

CI builds recently began failing on every GHC below 9.10 (8.8 – 9.8) while compiling a transitive dependency:

Text/Megaparsec/Stream.hs:517:26: error:
    • Variable not in scope: foldl' :: (b -> Char -> b) -> b -> [Char] -> b
    • Perhaps you want to add ‘foldl'’ to the import list
      in the import of ‘Data.Foldable’
Failed to build megaparsec-9.8.0 (which is required by test:test from
persistent-postgresql, test:test from persistent-sqlite and others).

megaparsec-9.8.0 uses an unqualified foldl', which is only re-exported from Prelude as of base-4.20 (GHC 9.10). On older compilers it's a hard "not in scope" compile error.

It surfaced now because the workflow runs cabal v2-freeze fresh on every run (there's no committed cabal.project.freeze), so the solver always floats to the newest Hackage release — and megaparsec-9.8.0 was recently published. GHC 9.10/9.12 compile it fine; everything ≤ 9.8 breaks. This affects master and every open PR equally, not any one branch.