[Inference] Let's use lowercase left/right truncation direction parameter (#1841)

First spotted by @alvarobartt on
[slack](https://huggingface.slack.com/archives/C064Z3GRU3U/p1762871481747849?thread_ts=1743434022.621759&cid=C064Z3GRU3U)
(internal).

The `FeatureExtractionInputTruncationDirection` parameter defined in the
feature-extraction specs currently have capitalized values `Left` /
`Right`. This comes from TEI implementation from which we pulled the
specs. Let's switch it to lowercase values for consistency with all
other specs.

In addition to this PR, we should:
- update TEI to accept `left`/`right` (not even sure we want backward
compatibility on this) => @alvarobartt would you be able to handle this?
- update `huggingface_hub` => I can take care of this
huggingface/huggingface_hub#3548
- update `hub-docs` => will happen automatically

I think it's fine to have breaking change for this. Moreover, specs /
docs are there more for information but are rarely enforced (expect in
TEI) so it shouldn't break anything client-side.

Author: Wauplin

packages/tasks/src/tasks/feature-extraction/inference.ts

@@ -38,4 +38,4 @@ export interface FeatureExtractionInput {
  * The text or list of texts to embed.
  */
 export type FeatureExtractionInputs = string[] | string;
-export type FeatureExtractionInputTruncationDirection = "Left" | "Right";
+export type FeatureExtractionInputTruncationDirection = "left" | "right";

packages/tasks/src/tasks/feature-extraction/spec/input.json

@@ -51,7 +51,7 @@
 	"$defs": {
 		"FeatureExtractionInputTruncationDirection": {
 			"type": "string",
-			"enum": ["Left", "Right"],
+			"enum": ["left", "right"],
 			"title": "FeatureExtractionInputTruncationDirection"
 		}
 	}