No easy way to reuse an uploaded file from its id

I've just implemented https://github.com/tidyverse/ellmer/pull/1119 aimed at improving a bit of UX in ellmer; here's the 🤖 summary of the equivalent in chatlas:

Once a file has been uploaded, reusing it in a later session means rebuilding the ContentUploaded yourself from what chat.files.get() returns:

from chatlas import ChatAnthropic, ContentUploaded

chat = ChatAnthropic()
meta = chat.files.get("file_011CRabc...")
report = ContentUploaded(id=meta.id, mime_type=meta.mime_type, provider=meta.provider)
chat.chat("Summarize the key findings.", report)

That works for Anthropic and Google, but fails for OpenAI because its file_get() sets mime_type=None and ContentUploaded.mime_type is a required str, so you have to guess the type from meta.filename yourself. ellmer is resolving this by having chat$file_get() return a ContentUploaded with the metadata on extra and guessing OpenAI's MIME type from the filename (tidyverse/ellmer#1116, tidyverse/ellmer#1119). Might be worth aligning, either by doing the same or by adding a FileMetadata.to_content() helper.

Start by tracing chat.files.get(), file_get(), FileMetadata, and ContentUploaded, then compare the equivalent ellmer changes in tidyverse/ellmer#1116 and #1119. Make a retrieved file reusable in a later chat, including suitable MIME metadata for OpenAI, or provide the proposed FileMetadata.to_content() helper; verify the existing Anthropic, Google, and OpenAI paths.