Retry "Exception: not primary" when writing

[thekid]

The clientside cluster view may become outdated, causing us to erroneously write to non-primary nodes. This is how this can occur (see #42 (comment)):

Event	Client view	Server side situation
Client: Before	(n/a)	{Primary: A, Secondary: [B, C]}
Client: Connect	{Primary: A, Secondary: [B, C]}	{Primary: A, Secondary: [B, C]}
Remote: Primary change	{Primary: A, Secondary: [B, C]}	{Primary: B, Secondary: [A, C]}
Client: Write	{Primary: A, Secondary: [B, C]}	{Primary: B, Secondary: [A, C]}

The client has an inconsistent view of the server side situation and uses node A for writing which it believes to be the primary, while in reality it's changed its role to a secondary.

If we encounter an error indicating this condition, we now refresh the cluster view, then retry the write once.

[thekid]

Released in https://github.com/xp-forge/mongodb/releases/tag/v2.1.0