Using cabal as a script interpreter doesn't write output to current working directory

I am trying to use cabal to run a one-off script which runs a plugin on the file and outputs some information to the current directory. However, the output gets directed somewhere else entirely.

For example:

```
#!/usr/bin/env cabal
{-# LANGUAGE DeriveGeneric , ViewPatterns, PatternSynonyms #-}
{- cabal:
build-depends: base, dump-core, process, criterion, deepseq
ghc-options: -fplugin=DumpCore -fplugin-opt DumpCore:core-html
-}
```

I want the output to be in a folder called `core-html` in the same place as the script. However, it doesn't end up there and I don't know where it ends up.

Reproduce the shebang-based cabal script example with the DumpCore plugin and observe where the core-html output is written. Start at cabal's script interpreter handling and trace how it determines the working directory; done means output is placed in a core-html folder beside the script, with coverage for this invocation.

Written by the indexing model from the issue text.