Add Data.Aeson.Decoding.Text, decodeStrictText :: Text -> ...

[phadej]

We avoid intermediate ByteString copy by not doing decode . TE.encodeUtf8, but instead working on Text value directly. As we know that the stream is valid Unicode (UTF8 or UTF16), we can also take some shortcuts.

One gotcha is that internal Text values (in Keys or Value Strings) most likely retain the original input Text value (its Array). It shouldn't be an issue if the Value is actually decoded so these Text values disapper, but if not (e.g. Object keys survive) then users might want to use Data.Text.copy.

With GHC-9.6.2 (text-2.0.2; UTF-8) the speedup is not huge, but noticeable anyway:

aeson/strict:                 OK (0.26s)
  462  μs ±  23 μs
aeson/text:                   OK (0.22s)
  399  μs ±  25 μs
aeson/text-via-bs:            OK (0.14s)
  473  μs ±  45 μs

With GHC-8.6.5 (text-1.2.3.0; UTF-16) the speedup is relatively more:

aeson/strict:                 OK (0.22s)
  819  μs ±  74 μs
aeson/text:                   OK (0.17s)
  593  μs ±  46 μs
aeson/text-via-bs:            OK (0.23s)
  875  μs ±  62 μs