ci: speed up test-coverage workflow with 4 CPUs, sharding, and path filtering

[randy3k]

Summary

This PR speeds up the test-coverage workflow through three complementary optimizations:

1. Parallel test execution with 4 vCPUs: Sets TESTTHAT_CPUS: 4 in the test-coverage workflow to take full advantage of standard GitHub Actions runners (4 vCPUs on Linux).
2. Path filtering: Adds paths-ignore for markdown documentation (**.md) and .gitignore to prevent unnecessary CI runs on doc-only changes.
3. 4-way balanced test sharding: Partitions all 61 test files across 4 balanced matrix shards running in parallel, uploading separate flags (shard-1, shard-2, shard-3, shard-4) to Codecov for complete aggregation:
   • Shard 1 (^[a-c]): 2,920 lines, 82 tests (contains completion)
   • Shard 2 (^[d-i]): 3,928 lines, 116 tests (definition, formatting, hover, index, etc.)
   • Shard 3 (^[j-r]): 3,144 lines, 100 tests (lintr, refactor, rename, references, etc.)
   • Shard 4 (^[s-z]): 3,494 lines, 101 tests (semantic-tokens, symbol, task, type-hierarchy, etc.)

Runtime Comparison

• Baseline (single thread, all tests): ~44 minutes
• 4 vCPUs (all tests): ~32 minutes
• 2-way naive sharding (unbalanced, shard 1 took 77% of assertions): ~30 minutes
• 4-way balanced sharding: expected ~8–10 minutes per shard

[renkun-ken]

LGTM Thanks!