S24.17: Collapse per-platform pool-size tunables to role-based tunables

[DavidCozens]

Parent epic: #254

Problem

Core/Interface/SolidSyslogTunablesDefaults.h carries 35 *_POOL_SIZE tunables. For platform/vendor-selected classes the same role is named once per implementation — SOLIDSYSLOG_POSIX_TCP_STREAM_POOL_SIZE, _WINSOCK_TCP_STREAM_, _PLUS_TCP_TCP_STREAM_, _LWIP_RAW_TCP_STREAM_ are four names for "how many TCP streams". Only one implementation is ever live in a given build (platform backends are mutually exclusive at link time), so:

• the integrator must know which platform they are on to find the knob that matters, and reasons about a role they cannot name;
• the table grows O(roles) with every new OS / network stack / crypto vendor instead of O(1).

SOLIDSYSLOG_ADDRESS_POOL_SIZE already proves the role-based model — one tunable shared across the Posix / Winsock / PlusTcp Address implementations. This story extends that established pattern consistently to the remaining platform- and vendor-selected classes.

Change

Collapse 8 role groups to single role-named tunables. Clean break — no deprecated aliases (pre-1.0).

New role tunable	Replaces	Default
SOLIDSYSLOG_TCP_STREAM_POOL_SIZE	POSIX / WINSOCK / PLUS_TCP / LWIP_RAW TCP stream	2
SOLIDSYSLOG_DATAGRAM_POOL_SIZE	POSIX / WINSOCK / PLUS_TCP / LWIP_RAW datagram	1
SOLIDSYSLOG_RESOLVER_POOL_SIZE	GETADDRINFO / WINSOCK / PLUS_TCP / LWIP_RAW / LWIP_RAW_DNS	1
SOLIDSYSLOG_MUTEX_POOL_SIZE	POSIX / WINDOWS / FREE_RTOS mutex	1
SOLIDSYSLOG_FILE_POOL_SIZE	POSIX / WINDOWS / FATFS file	1
SOLIDSYSLOG_ATOMIC_COUNTER_POOL_SIZE	STD / WINDOWS atomic counter	1
SOLIDSYSLOG_TLS_STREAM_POOL_SIZE	OpenSSL / mbedTLS TLS stream (name already exists for OpenSSL; mbedTLS folds in)	1
SOLIDSYSLOG_HMAC_SHA256_POLICY_POOL_SIZE	OpenSSL / mbedTLS HMAC policy	1

Net: 35 -> 18 pool tunables.

Within every role group the current defaults are already uniform (all TCP streams = 2; all datagrams / resolvers / mutexes / files / counters / TLS / HMAC = 1), so there is no default-value reconciliation — each role macro simply takes the group common value. Preserve all floor static_asserts.

Convention (the new rule)

Documented at the head of the tunables file and codified in docs/NAMING.md:

Pool-size tunables are named by role, not platform. The pool counts what the integrator creates, regardless of which platform/vendor implementation provides it.

Two same-platform coexistence cases now share a role pool (instead of two pools of 1):

• lwIP numeric + DNS resolver — both compile on an lwIP target and may coexist (DNS is a superset). Wire both -> set SOLIDSYSLOG_RESOLVER_POOL_SIZE to 2.
• Two crypto vendors in one binary (OpenSSL + mbedTLS TLS stream, or both HMAC policies) — one vendor per build is the norm, but if both are wired, set the role pool accordingly.

This matches how Sender / Buffer pools already behave.

Files affected

One mechanical macro reference per file, no logic change:

• Core/Interface/SolidSyslogTunablesDefaults.h — consolidate 25 macro blocks into 8 role blocks (keep floor static_asserts).
• ~25 Platform/**/*Static.c — point each pool array at its role macro.
• ~25 Tests/** — reference the role macro (pool/exhaustion tests).
• Bdd/Targets/FreeRtos/solidsyslog_user_tunables.h, Bdd/Targets/FreeRtosLwip/solidsyslog_user_tunables.h — FREE_RTOS_MUTEX_POOL_SIZE -> MUTEX_POOL_SIZE.
• Tests/Fixtures/SmallMessageSizeTunables.h — two atomic-counter override lines collapse to one SOLIDSYSLOG_ATOMIC_COUNTER_POOL_SIZE.
• CLAUDE.md, docs/NAMING.md, docs/integrating-lwip.md, docs/misra-deviations.md — name updates + convention.

No cross-macro value conflicts exist in the tree (verified): the only multi-override sites already set the same value for siblings that merge.

Test evidence

• debug + clang-debug green.
• FreeRTOS-host, lwIP, and mbedTLS local images green (touches all platform trees).
• No coverage delta — rename / consolidation only, no new branches.

Acceptance criteria

• 35 pool tunables reduced to 18; the 17 removed names appear nowhere in the tree.
• All 8 role tunables retain a floor static_assert.
• Role-not-platform convention documented in the tunables header and docs/NAMING.md.
• All platform CI lanes green.

[DavidCozens]

Coexistence audit (pre-implementation)

Verified that no single BDD executable creates two implementations of the same role — so the role collapse needs no summed overrides and no new tunables. Traced actual _Create wiring in every target:

Target	TCP stream	Datagram	Resolver	TLS	HMAC	Mutex
Linux	PosixTcp x2 (plain + TLS-underlying)	Posix x1	one (shared)	OpenSSL x1	OpenSSL	—
Windows	WinsockTcp x2	Winsock x1	one	OpenSSL x1	OpenSSL	Windows
FreeRtos	PlusTcpTcp x2	PlusTcp x1	PlusTcp x1	mbedTLS x1	mbedTLS x1	FreeRtos x2
FreeRtosLwip	LwipRawTcp x2	LwipRaw x1	LwipRawDns x1	mbedTLS x1	—	FreeRtos x2

Key findings:

• lwIP resolver coexistence does not occur — FreeRtosLwip creates only SolidSyslogLwipRawDnsResolver (shared across all senders); the numeric SolidSyslogLwipRawResolver is never created in any target. RESOLVER count = 1.
• No dual crypto vendor — hosted = OpenSSL-only, embedded = mbedTLS-only. tls/mtls are a runtime mode of one TLS slot (BddTargetTlsSender_Create(resolver, mtlsModeActive)), not two concurrent senders. TLS_STREAM = 1, TCP_STREAM = 2.
• SOLIDSYSLOG_FREERTOS_NET=BOTH compiles both PlusTcp + LwipRaw *Static.c but the built executable only _Creates the PlusTcp set; the unused backend pool is static slack, no conflict after collapse.

Override mapping is 1:1, no sums:

• FREE_RTOS_MUTEX_POOL_SIZE 2U -> MUTEX_POOL_SIZE 2U (both FreeRTOS targets)
• STD_ATOMIC_COUNTER 2U + WINDOWS_ATOMIC_COUNTER 2U -> single ATOMIC_COUNTER_POOL_SIZE 2U (Tests/Fixtures/SmallMessageSizeTunables.h)

Forward guard (add to the convention doc)

If a future target ever wires two implementations of the same role in one executable (e.g. a numeric + DNS resolver, or two crypto vendors), its role pool must be sized to the sum of concurrent instances. The role pool counts instances, not implementations.