std.fs.File: fix positional/streaming seekBy

[rhizoome]

commit 1: seekBy should use the logical position, but it used the file position which includes the buffered data. the test did not catch this because it used an empty buffer. fixes #25087 #25020

commit 2: streaming seekBy completely ignored the buffer. the new logic checks if the seek offset is smaller than the buffer and adds the offset to r.interface.seek. otherwise the old logic is invoked.

[moosichu]

The main thing I would say reading through the PR, is that the tests seem to be testing the internal behaviour of the reader implementation (how the buffering works etc.) depending on modes and I don't think that's super important.

Ideally they would be testing that actual API boundary/contract. Which is that when seeking, the logical pos in the file changes as expected and that the data read through calls to the reader API return the data that is expected.

So instead of tests like:

    // seek within bufferedLen
    const pos_within = reader.pos;
    try reader.seekBy(2);
    // wasi does not buffer in this case
    if (builtin.target.os.tag != .wasi) {
        try testing.expectEqual(pos_within, reader.pos);
    }

I personally would test for:

    const logical_pos_before = reader.logicalPos();
    try reader.seekBy(2);
    try testing.expectEqual(logical_pos_before + 2, reader.logicalPos());

(and then that the buffer returns the correct data, but that is already being done)

Maybe testing that these all work based on seeking within and beyond the buffer could still be useful as they invoke different internal logic. But I don't think checking that we get specific values into the inteferace seek and end fields is super useful (for example).

Also adding a negative seekBy to the test could be useful as well for: #25020

This is all just my opinion though so take it with a grain of salt (as others might have different thoughts)! And I hope it makes sense as I went on a couple tangents.

[rhizoome]

Updated tests to not test internals, just the interface. I also added a negative seekBy. This was the version of the tests that made sure, that the buffer is discarded. I am now pretty sure that testing internals was a bad idea.