Support buffering by bytes rather than number of lines

[ccampo133]

Currently the message buffer size represents the number of log messages (lines) and not their individual sizes. This is desirable in many (most?) cases, however this may not be so useful when log messages are extremely large, as it could result in log2fluent consuming lots of memory, or worse, stalling, while it attempts to read the long lines into memory.

The buffering strategy should be configurable by the user via some flag(s), to be either line or byte based. This will require a different implementation of the reader goroutine.