fix(internal/utils): keep byteReader position after failed seeks

[fallintoplace]

byteReader updated its cached position even when the underlying seek failed, which could make Peek and Read disagree about where the reader was. Discard also returned io.EOF when it consumed exactly the remaining bytes.

This only updates the cached position after a successful seek and treats exact exhaustion as a successful discard. Added focused regressions for both behaviors.

Tests: go test ./internal/utils

[zeroshade]

Verified both fixes against current main:

1. Seek unconditionally did r.pos = int(pos). Since bytes.Reader.Seek returns pos = 0 on failure (invalid whence or negative resulting position), a failed seek corrupted the tracked position to 0 while the underlying reader stayed put — leaving Peek and Read disagreeing about the location. Guarding on err == nil is correct.

2. Discard used newPos >= len(r.buf), so discarding exactly the remaining bytes reported io.EOF even though all n bytes were successfully discarded. > is right and matches both stdlib bufio.Reader.Discard semantics and the sibling bufferedReader.Discard in this same file.

No regression risk on the EOF change: the only caller (page_reader.go) ignores Discard's return values. Both tests distinguish pre- from post-fix behavior. LGTM.