Bug fixes and small refactoring for carbonserver and trie index

[bom-d-van]

It's mainly a misc changes and bug fixes.

Mostly backward-compatible changes except the default values of max-metrics-globbed and max-metrics-rendered.

1. cache-scan: only init Cache.Shard.adds when the feature is enabled

cache-scan is a feature in go-cabron to enable quicker visibility of new metrics.
However, if the feature is not enabled, no need to cache the new metrics info
in Shard.adds map.

2. trie: numerious optimization and bug fixes for concurrent and realtime index

• Fix incorrect index total_runtime calculation
• Stop caching new metrics between file list scanning (which mean file list scanning interval should be slower than new metrics flushing to disk)
• Start new file list scanning on startup right after loading files from cache: to avoid serving old file list cache too long
• Move down the trie_count_nodes_time codes down to be within the indexing "block"
• Add a new counter cache.droppedRealtimeIndex (use camelCase to stay consistent with the rest of cache.* metrics)
• Re-initiating trieIndex when failed to load metrics from cache file
• Some other minor changes here and there

3. carbonserver: change max-metrics-globbed and max-metrics-rendered defaults (for issue: [FR] Returns error or generating error log when render and glob throttling kicks in #414)

The default values might be too low for some of the users and it is a default introduced mainly
for trie index. This commit change the default to a much bigger value to avoid silently truncated
query result.

4. carbonserver: change error/debug log destination to access.log for CarbonserverListener.prepareDataProto