Add subgraph name to tlparse for regional inductor

[yushangdi]

Summary

• Add support for the subgraph_name field in PyTorch structured logs (used by regional inductor for repeated subgraphs, partitioned forward/backward subgraphs, etc.)
• When subgraph_name is present on a log entry, the HTML index renders a "Subgraph Name: ..." group header to visually separate artifacts by subgraph
• Files remain flat in the compile ID directory (no new subdirectories)

Changes

src/types.rs — Add subgraph_name: Option<String> to Envelope so serde deserializes the new field. Add starts_group and group_name to OutputFile for rendering group separators.

src/lib.rs — Thread the envelope's subgraph_name through add_file_output into each OutputFile. After all parsing, compute group boundaries by detecting when consecutive files have different group_name values and setting starts_group = true.

src/parsers.rs — Copy the two new OutputFile fields in the CompilationMetricsParser output remapping.

src/templates.rs — Render a <li> group header with top margin when starts_group is true.

tests/inputs/subgraph_name.log — Minimal 6-line test log with 3 artifacts: one with no subgraph, one with repeated_subgraph0, one with partitioned_bw_subgraph_1_0.

tests/integration_test.rs — test_subgraph_name_grouping: verifies files are flat (no subgraph subdirectories), and the index HTML contains the expected group headers.

Test plan

• cargo test — all 64 tests pass (63 existing + 1 new)
• Manual run on a real log with subgraph_name entries confirms group headers appear in index.html

Example output:

For annotation based regional inductor:

For invoke_subgraph based regional inductor: