chore: re-measure every resource-cost claim, and regenerate the examples that produce them

[DavidCozens]

Every claim the project makes about flash, RAM and stack cost is measured from the
two example stacks. Those stacks are stale, and #750 has moved the default they
were measured against, so the figures the release would ship are wrong. This is the
last story before #650 cuts 0.1.0, so that the release carries measured sizes.

The examples are the instrument

solid-syslog-example (CMake, 15 commits) and solid-syslog-example-make (Make,
22 commits) teach integration one commit at a time, each stage measured. The
history is the artefact, so an early commit cannot be patched without rewriting
every SHA after it. Both are regenerated as a stack rather than fixed piecemeal.

Neither is broken today: each pins the library (e4bc54d via FetchContent,
0d212aa via submodule), so both still build. They demonstrate an old API against
an old pin - stale, not failing. Nothing in docs/, the README or the website
cites an example commit SHA, so the rewrite breaks no inbound link.

Sequencing

1. Regenerate both stacks on current main, measuring each stage.
2. Correct every cost claim below from those measurements, and re-derive the
   sizing narrative described under Not just the figures.
3. Cut 0.1.0 release #650 cuts 0.1.0, shipping the corrected docs.
4. Re-pin both stacks to the release tag if the cut moved the SHA. The release
   commit changes no library source, so the figures should stand - confirm rather
   than re-measure.

Carry into the regeneration

Three breaking refactors land between the pins and main. Each fails to compile,
deliberately - no compatibility aliases were left:

• refactor!: make every platform class carry its registry token #721 - every platform class carries its registry token.
• refactor!: prefix every public error constant with SOLIDSYSLOG_ #724 - every public error constant prefixed SOLIDSYSLOG_.
• refactor!: S23.22 name platform callbacks for what they supply #739 - platform callback providers renamed (SolidSyslog<Plat>Hostname_Get
  to SolidSyslog<Plat>_GetHostname, and the same shape for GetProcessId,
  GetTimestamp, GetSysUpTime, Sleep). Header file names are unchanged.

Also:

• fix: give the config lock a context and require it to tolerate blocking #758 - SolidSyslog_SetConfigLock gained a void* context, and the lock
  must tolerate blocking.
• Default SOLIDSYSLOG_MAX_MESSAGE_SIZE to 480, the length every receiver must accept #750 - the default SOLIDSYSLOG_MAX_MESSAGE_SIZE is 480.
• Licensing (docs: offer SolidSyslog under three alternative licences #761, docs: explain the licence expression and make its LicenseRefs resolvable #762, docs: add a licence-header gate, then stamp the shipped sources #763). solid-syslog-example-make's
  third_party/README.md calls the library PolyForm Noncommercial 1.0.0, which is
  stale; solid-syslog-example says nothing about the library's licence at all.
  Both repos remain 0BSD in their own right. The submodule picks up the corrected
  LICENSE.md and LICENSES/ once the pin moves; only the prose needs a human.
• Every shipped source file now carries an SPDX header (docs: add a licence-header gate, then stamp the shipped sources #763), so copied sources
  in an example show it.

Resource-cost claims

Nothing here is adjusted on paper. The known inventory:

• README.md:14-16 - the headline figures: +5 KB flash / 0.4 KB RAM for a
  record on the wire, +13.5 KB flash / 37 KB RAM for the whole path.
• docs/hardening-path.md - a **Cost.** line on each of the twenty stages,
  plus the reconciliation paragraph quoting 13.5 KB against the ~13.9 KB the
  column sums to.
• The example repositories themselves, which carry the exact figures, the diff
  that produced each, and the reasoning behind it.
• The website, which is outside this repository and needs checking against the
  same measurements rather than assumed consistent.

Not just the figures

Stages 7, 13 and 14 of docs/hardening-path.md are a sizing narrative - take
the default, hit truncation at a 256-byte cap, double it to 512, then shrink the
ring - and it is written against a 2048-byte default. At 480 the premise of each
stage changes: stage 7 is no longer mainly a reduction, and stage 13's "double the
cap to 512" is a slight raise above the default rather than a doubling. Re-derive
the narrative, then measure it.

Done when

• Both stacks rebuild on main, commit by commit.
• Every figure above is measured at the new default, and the sizing narrative
  follows from the default the library now ships.
• The website carries no cost claim that disagrees with the measurements.
• The licence prose in both repos states the library's licence correctly.
• After the cut, both stacks pin the release and their figures are confirmed.

[DavidCozens]

Done, apart from step 4, which is dropped deliberately.

1. Both stacks regenerated on current main, measuring each stage. Each is now the
Baseline commit plus 19 commits, one merged pull request per stage, linear and CI green,
pinned at a8d3979. The old histories are tagged pre-480-regen on both remotes. The
three breaking refactors carried in as planned, and the licensing prose was corrected
with them. Every figure is measured by the device itself; the published pair is Flash
+13,788 B, RAM +36,048 B for the whole path, identical from both stacks. Per-stage
deltas agree between the repos to within 16 bytes of flash and 4 of RAM, which is
padding rather than divergence — app/ is byte-identical across them.

2. Every cost claim corrected, and the sizing narrative re-derived. #769 re-measured
docs/hardening-path.md and the README headline pair; #771 corrected the meta
description that had survived the restructure. At the 480-byte default the message cap
never needs raising, so the two sizing stages are gone and a right-sizing stage joins at
the end: twenty stages become eighteen. Nine stages measured within their published
figure and were left untouched. Two stale claims turned up alongside — the baseline
builds at C17, and the handler prints one bad-config line per missing collaborator — and
review of the change surfaced #770, now milestoned 0.2.0.

3. #650 cuts 0.1.0 with the corrected docs. The website was updated and published on
2026-08-17.

4. Not doing the re-pin. The figures are a rough example, each repo states its pin
plainly, and the release commit changes no library source — so re-pinning would cost two
history rewrites and buy nothing.