Remove the duplicate generateProjectionSchema and timestamp-millis helpers

#16639 wrote down where schema helpers belong and removed the duplication inside the Avro/HoodieSchema util classes themselves. Two look-alike pairs are still standing outside them, and both are the failure mode that issue describes: the same logic re-implemented because the existing one was not found.

### 1. `generateProjectionSchema`, implemented twice

- `HoodieSchemaUtils#generateProjectionSchema(HoodieSchema, List)` -- `hudi-common/src/main/java/org/apache/hudi/common/schema/HoodieSchemaUtils.java:496`
- `HoodieRealtimeRecordReaderUtils#generateProjectionSchema(HoodieSchema, Map, List)` -- `hudi-hadoop-mr/src/main/java/org/apache/hudi/hadoop/utils/HoodieRealtimeRecordReaderUtils.java:124`

Same loop, same case-insensitive lookup, same exception down to the punctuation:

```
"Field " + fn + " not found in log schema. Query cannot proceed! Derived Schema Fields: "
```

The wording shows the copy direction: the general-purpose hudi-common helper still says "log schema", which only makes sense in the realtime-reader caller.

Two differences a merge has to carry over:

- The hadoop-mr overload takes a prebuilt `schemaFieldsMap` rather than building one. Its sole caller, `AbstractRealtimeRecordReader:186`, builds that map at `:182` via `getNameToFieldMap(writerSchema)` and also hands it to `constructHiveOrderedSchema` at `:183`, so the map is genuinely shared and worth keeping out of the helper. `getNameToFieldMap` lowercases its keys exactly as the hudi-common version does inline.
- The hadoop-mr version calls the 5-arg `HoodieSchema.createRecord` and propagates `writeSchema.isError()`. The hudi-common version calls the 4-arg overload, which hard-codes `isError = false`.

Suggested shape: add a `HoodieSchemaUtils#generateProjectionSchema` overload that accepts the prebuilt map and preserves `isError`, delete the hadoop-mr implementation, and point `AbstractRealtimeRecordReader` at the survivor.

### 2. The timestamp-millis check, split across two domains

- `HoodieSchemaRepair#hasTimestampMillisField(HoodieSchema)` -- `hudi-common/src/main/java/org/apache/hudi/common/schema/HoodieSchemaRepair.java:232`. Recursive over a whole table schema (RECORD, ARRAY, MAP, UNION), true if any leaf is a millis timestamp.
- `HoodieTableMetadataUtil#isTimestampMillisField(HoodieSchema)` -- `hudi-common/src/main/java/org/apache/hudi/metadata/HoodieTableMetadataUtil.java:506`. The same leaf test for a single field schema, after `getNonNullType()`.

The leaf test is what is duplicated: unwrap the union, check `TIMESTAMP` with `TimePrecision.MILLIS`. The `HoodieSchemaRepair` class javadoc already cross-references the sibling, so the relationship is known; the walker should call the single-field predicate instead of re-implementing it, leaving one definition of "is a millis timestamp".

### Scope

Both are `refactor(schema)` and behavior-preserving provided the two differences in part 1 are carried over. Existing coverage: `TestHoodieSchemaUtils`, `TestHoodieRealtimeRecordReaderUtils` and `TestHoodieRealtimeRecordReader` for part 1; `TestHoodieSchemaRepair` and `TestHoodieTableMetadataUtil` for part 2.

Related: #16639 (the routing rule and the remaining Avro round-trips).