v0.1.7 fs-bench-pro pilot 2: complete SDK edit-* families

[yifanxuaaa]

Parent: #230. Prerequisite pilot 2 of 3. Do not begin migration implementation or collection for the remaining fs-bench-pro families until all three pilots have complete PASS evidence. Opening this issue approves no numeric performance result.

Scope

Migrate the complete active SDK edit-* cluster as one issue and one authentic edit route:

Family	Active cases	Operations
edit_length_preserving	12	4 KiB head/middle/tail overwrite across 1/10/100/500 MiB
edit_length_changing	32	Insert/delete/append/prepend/grow/shrink/truncate/zero-extend across those tiers
edit_canonical_chunk_count	12	64 KiB replacement preserving/increasing/decreasing canonical chunk count across those tiers

Total: 56 active cases. The five old edit_length_changing_capped v1 rows are historical optional duplicates; the active capped-v2 replacements are already among the 32 length-changing cases. Do not report 61 active rows.

Product and benchmark gates

• Add and verify the public authenticated host SDK/control operation that reaches the live v0.1.7 Workspace range-edit API. Each registered edit uses Client::edit_workspace_file_range or Client::edit_workspace_file_ranges as declared, followed by its explicit Commit. The v0.1.7 co-design pair 1: projection and runtime — FUSE against the workspace accumulator #179 mounted POSIX write path, direct Service EditFile, and Stage 6 C1 edits cannot substitute for SDK edit evidence.
• Prove one intended SDK edit call and the registered Commit topology, zero edit-caused FUSE WRITE, mounted visibility, exact bytes/chunks, retained old roots, and independent reopened-state verification. Do not turn an SDK edit into a shell copy/rename or a direct Store mutation.
• Freeze all 56 case identities, fixtures, seed, operation/timer boundaries, cache state, resource and correctness gates, and prospective numeric limits before code/collection. Do not import old five-repetition statistics or silently reuse latency limits for a changed v0.1.7 route.
• Iterate with one selected case at a time, then collect exactly one eligible performance sample per registered case/arm on the final source, followed by separate identity-matched verification and cleanup. Keep all failures and unrun rows; enforce complete-command budgets without shrinking workloads, warming sources, or hiding setup inside a different phase.

Completion gate

All 56 active cases have eligible performance PASS, independent verification PASS, cleanup PASS, and complete route/cache/source custody. The optional capped-v1 rows remain explicitly historical. A local Workspace test, direct C1 edit, or one successful tier does not clear this pilot.

Source references: docs/roadmap/0.1/benchmarking.md, docs/general/benchmark_rules.md (SDK file-edit invariant), and core/docs/architecture/proposal/fuse-workspace-snapshot-overlay/06-benchmark-qualification-map.md.