Column Seeking Rework

This PR reworks the split implementation we have for column seeking in NpgsqlDataReader.

Where previously we had two methods split by behavior; one for non buffered mode (sync only) and one for sequential mode (async/sync).

We now just have a sync and an async implementation, with the sync method having a tiny bit of additional validation for sequential behavior.

Fast paths should essentially be unaffected and we might even see a small speedup in TE non sequential scenarios due to streamlining column seeking but also PgReader.{Init,Commit} even more.

For the async implementation we're exploiting the fact that any rereading or rewinding necessarily means we already have the data. So we're able to fully delegate all that work to the sync implementation.

Similarly we're now always using the sync implementation if we know the row is buffered entirely.

Async Seeking Caveat

One optimization that I introduced in 8.0 that didn't make it over is when doing an *async sequential* seek (so mostly via IsDBNullAsync or GetFieldValueAsync). We used to try and seek up to the column that was buffered before switching to the async method, saving a small bit of time setting up the state machine and entering it.

In my opinion that complexity isn't worth carrying over. It was initially introduced because sequential seeking had no sync only method at all, impacting the super common sync methods like GetFieldValue. While the async column reading methods are rarely used. Even then the new implementation will still use the sync path when the entire row is buffered.

On the other hand *sync sequential* seeking now uses the same heavily optimized implementation as buffered mode does.