Distributed node-level and edge-level temporal sampling for hetero

[kgajdamo]

The purpose of this PR is to enable distributed sampling with node-level and edge-level temporal information for heterogeneous graphs.

Description:

• Heterogeneous temporal sampling is analogous to homogeneous temporal sampling, but takes into account the presence of node types and edge types. We define the node time information for each node type and the edge time information for each edge type.
• Because of the lack of the node_store/edge_store in the feature store we determine whether to use node-level or edge-level temporal sampling based on the time_attr value ('time' or 'edge_time').
• seed_time is mandatory for edge-level sampling.
• seed_time field has been added to the NodeDict class to store the time information of the source nodes.
• The time information of source nodes for the next layer is calculated based on the subgraph ID a given node belongs to (each subgraph has a seed time specified, which is common for all source nodes).

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (d1f305b) 89.49% compared to head (215a0d9) 89.50%.

❗ Current head 215a0d9 differs from pull request most recent head 1ad465a. Consider uploading reports for the commit 1ad465a to get more accurate results

Files	Patch %	Lines
torch_geometric/sampler/neighbor_sampler.py	90.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8624      +/-   ##
==========================================
+ Coverage   89.49%   89.50%   +0.01%     
==========================================
  Files         481      481              
  Lines       31084    31096      +12     
==========================================
+ Hits        27819    27833      +14     
+ Misses       3265     3263       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.