Add a more descriptive error for text/plain files that are not UTF-8

[handsomefox]

Description of the bug:

I've had to recently debug an issue .txt file. We had no conversions to UTF-8 inplace and this resulted in a confusing problem, where Gemini uploads just result in:

googleapi: got HTTP response code 400 with body: [{\n  \"error\": {\n    \"code\": 400,\n    \"message\": \"Request contains an invalid argument.\",\n    \"status\": \"INVALID_ARGUMENT\"\n  }\n}\n]"

This pretty much gives you nothing and I've spent some time to debug.

I think two things can fix this issue for future users:

1. The docs for client.UploadFile can better indicate that UTF-8 is required for text files, or validate it itself.
2. The API error that we get should include more details about happened.

The validation in the first point is quite easy if you decode the content-type correctly, since anything that begins with text/ you can just call utf8.Valid(b) on.
The second point I have no idea about.

You could also provide a conversion method for generic cases like UTF16(LE/BE) and Windows 1252/1251 since they're quite common.

Actual vs expected behavior:

No response

Any other information you'd like to share?

No response

[eliben]

We're trying not to add validation around the backend model, but we could perhaps document this better

@jba WDYT?

[jba]

We should document that UTF-8 is required, and add tests to make sure that it is indeed broken for non-UTF-8. That way if the backend learns non-UTF-8, we will know to remove the doc.

[eliben]

@handsomefox can you link an example of a text file encoded in 1252 or whatever you tried, that gave you this error?
I sent #224 with a test for a generic bad encoding, which seems to result in a different error from the service

[handsomefox]

@eliben i've retested all of these files and they've all returned [{\n \"error\": {\n \"code\": 400,\n \"message\": \"Request contains an invalid argument.\",\n \"status\": \"INVALID_ARGUMENT\"\n }\n}\n] and no errors when encoded to utf8

utf16le.txt
1251.txt
1252.txt

From the differences from your test, I can only see that I'm using ChatSession.SendMessage instead of Model.GenerateContent, some model parameters are changed, system instruction is added and the file comes before the text but that doesn't seem like it should matter.

[eliben]

I've sent #228 adding one of the files @handsomefox uploaded, it indeed errors out as described.

W.r.t. updating the documentation - the natural place for this seems to be https://pkg.go.dev/github.com/google/generative-ai-go/genai#UploadFileOptions, but it already documents the MIME pointing to a link that was redirected and now doesn't seem to contain the required information. I think we should find where the supported MIME types are described now.