feat(casing): snake/wire surface parity across params, bodies, presentation and columns (#131); hyphenated wire names alias cleanly (#119)

[jeffreyaven]

Closes #131. Closes #119.

Every change is gated on provider doc config (request.nativeCasing on the
method, config.snake_case_aliases on the provider). Resolution is wire-first
on every surface: the key as written matches first, the snake alias second, and
no wire spelling is removed from any resolvable set. Absent both flags every
path is byte-identical.

#119 - hyphenated wire names

casing.ToSnake now treats - as a word boundary, transforming each segment
with the existing botocore rules: openai-organization -> openai_organization,
OpenAI-Organization -> open_ai_organization, X-Amz-Date -> x_amz_date.
Non-hyphenated names are unchanged (VPCId -> vpc_id, training_file ->
identity). FromSnake(_, kebab) was already the correct inverse.

#131

item	change
1. column surface	Schema.GetProperty and FindByPath accept the snake alias of a wire property when snake_case_aliases is on (wire-first), so snake projections resolve their property schema.
2. acronym round-trip	GetParameter, parameterMatch, namespaceParameterMatch and GetOperationParameter resolve a snake key through the declared wire names (ToSnake-derived alias map) before the mechanical FromSnake fallback, so ip_protocol -> IPProtocol, binary_id -> BinaryId. GetParametersIncludingNativeCasing aliases data__ body keys with the prefix kept.
3. default data__ regime	revertRequestBodyAttributeRename maps a snake body key to its wire property when the method declares nativeCasing, so data__storage_class and data__storageClass both send {"storageClass": ...}. Unknown keys pass through as today (an error here would break free-form bodies).
4. presentation	RenameRequestBodyAttribute (SHOW INSERT) and ToPresentationMap (SHOW METHODS) present snake names only when the method declares nativeCasing and the provider enables snake_case_aliases; nested body contents and server variables keep wire casing. The unexported rename used for resolution is unchanged, so GetParameters() keeps the wire keys.
5. body-less EXEC	GetRequestBodySchema returns nil, nil when no body schema is declared, so a metadata-only request block on a body-less method no longer fails before dispatch. The internal body-attribute accessors keep their error contract, which preserves routing exactly (they previously short-circuited before required server variables were added).

Verification

• go build ./..., go test ./... (21 packages), gofmt on changed hunks, lint
  clean on changed lines. New tests: pkg/casing hyphen cases and
  internal/anysdk/casing_alias_resolution_test.go (one test per item, each
  asserting the wire spelling still resolves and the flag-off path is unchanged).
• End-to-end: stackql built against this branch vs v0.5.4-alpha01, local file
  registry provider with snake_case_aliases: true and per-method
  nativeCasing, mock logging method/path/query/header/body:

statement	v0.5.4-alpha01	this branch
WHERE openai_organization = 'org-x' (kebab header)	could not locate symbol	header openai-organization: org-x on the wire
INSERT ... (data__name, data__storage_class)	body {"storage_class": ...} (silently ignored by an API)	body {"storageClass": ...}
INSERT ... (data__name, data__storageClass)	{"storageClass": ...}	unchanged
EXEC ...stop @machineId (body-less, request: {nativeCasing})	no request body for operation	dispatched POST /machines/m1/stop
SHOW METHODS (stop)	machineId	machine_id
SHOW INSERT	data__storageClass	data__storage_class
WHERE max_results = 2 / WHERE maxResults = 2	?maxResults=2	unchanged
SELECT machine_type, creation_timestamp, ip_protocol	rows	unchanged

Remaining in stackql (wire-through)

Two acceptance points in #131 sit in stackql's own layers and are unchanged by
this PR; both now have the any-sdk primitives they need:

• Wire-spelled column under the flag (SELECT machineType with
  snake_case_aliases): still no such column: machineType from the SQL
  backend. unary_selection.go builds the column descriptor from the SQL
  identifier; with FindByPath now alias-aware it can rename the descriptor to
  the resolved property's snake display name.
• Required EXEC argument as snake (EXEC ... @machine_id): the analyzer
  matches required exec args by wire name; resolving them through
  GetParameter / GetParametersIncludingNativeCasing makes required and
  optional exec args uniform.
• analyzeSchemaVsMap should nil-guard the schema now that a body-less method
  reports nil, nil: an EXEC that supplies a payload to such a method must
  error rather than dereference nil.

🤖 Generated with Claude Code

[general-kroll-4-life]

if a nil requestBodySchema is an error state and echemaErr is prior nil, then should update this to some hardcoded error here

[jeffreyaven]

Agreed. It now returns noRequestBodyError() when the schema is nil with no prior error, matching the other body-attribute accessors, and the test asserts the error. Fixed in 07de24e.

[general-kroll-4-life]

just the one thing. The early return in getRequestBodyStringifiedPaths looks to be an error state. If so, then the error should be populated if empty. If not, we can go ahead as is. I prefer minimalism so dont want splitting or commments