Optimizing multivariate metrics representation to provide native support within the client SDK

[lquerel]

Multivariate metrics already compress quite efficiently due to the grouping of duplicated attributes among metrics. Currently, each type of data point (sum, gauge, histogram, etc.) has a dedicated attribute table/record.

For future client SDKs that natively support multivariate metrics, it should be possible to eliminate the in-memory duplication of attributes. This will necessitate the creation of a single DATA_POINT_ATTRS table that can be shared among different types of data points.

Sub-tasks:

• Rename the Arrow Buffer METRICS to UNIVARIATE_METRICS to prepare the creation of a future MULTIVARIATE_METRICS buffer type. #19
• Implement the logic around the MULTIVARIATE_METRICS buffer type.