docs(hardware): record TTSKY26b withdrawal in silicon_anchor §3

[dmitrii-f-t27]

README.md:180 and :304 both cite docs/hardware/silicon_anchor.md §3.2 as their evidence for WITHDRAWN. A reviewer following that citation lands on the opposite claim: a grep for withdraw/refund/cancel in silicon_anchor.md returns zero matches, §3.2 marks all three crowns Submitted, §3.3 plans validation when chips return (expected ~Nov 2026), and §3.4 asserts the README's TTSKY26b claim is correct.

§3 now records the withdrawal: submitted 2026-05-17, cancelled and refunded, no die exists, no silicon route selected. §3.3 is retitled as a plan that cannot be executed and the date is removed. §3.4 is corrected. The wording follows README.md:39 and :180, which are already right, so the repo says one thing.

Separately, docs/hardware/README.md:28 cites gf16_benchmarks.json as evidence for a verified row. That file does not exist anywhere in the repo. The row now says so, and names what does exist — gf16_bench_results.json, a software benchmark (MSE, ns/op) with no FPGA frequency and no RTL test count.

I did not quietly substitute it: it is not hardware evidence and should not read as if it were. If a real hardware artefact exists elsewhere, point the row at that instead.

[dmitrii-f-t27]

@gHashTag — this is one of six PRs from a documentation and CI audit across the Trinity repos. Requesting your review here; I do not have permission to add you as a reviewer on this repo, hence the mention.

The full set:

• docs(readme): refresh three stale Metrics rows from their own commands tri-net#380 — three stale Metrics rows re-derived from the commands printed in their own rows
• Unbreak FPGA CI and correct two unbacked hardware claims t27#2081 — unbreak fpga-synthesis and fpga-bitstream, drop GREEN from three red README rows, remove two unbacked hardware figures
• fix(ci): make the three document gates actually run trinity-fpga#542 — make three document gates actually run (all three were vacuous or crashing)
• docs: reconcile catalog counts, LUT figures and device manifest against their sources trinity-fpga#543 — reconcile catalog counts, LUT figures and the device manifest against their sources
• docs(hardware): record TTSKY26b withdrawal in silicon_anchor §3 #52 — this one
• docs: remove TT SKY26b delivery date and mainnet schedule trinity-contracts#2 — remove the TT SKY26b delivery date and the mainnet schedule

Each PR body states what it does not fix as well as what it does. They are independent — merge or reject them separately.