fix(schema-compiler): use ClickHouse String type in Tesseract casts (#10316)

[igorlukanin]

Issue

Fixes the ClickHouse string-cast casing bug reported in #10316.

When the Tesseract SQL planner (CUBEJS_TESSERACT_SQL_PLANNER=true) renders a string cast on ClickHouse — e.g. the composite-primary-key concatenation behind a count measure queried through a join — it emits CAST(x AS STRING). STRING is not a valid ClickHouse type name (ClickHouse uses String, case-sensitive), so the query is rejected.

-- Tesseract (buggy):
count(CAST(order_id AS STRING) || CAST(product_id AS STRING))
-- Legacy planner (correct):
count(toString(CAST(`order_id` as String)) || toString(CAST(`product_id` as String)))

Root cause

BaseQuery defines types.string = 'STRING'. The legacy ClickHouse path never hits it because ClickHouseQuery.castToString() is overridden to emit String. Tesseract, however, renders the types.string template directly, and ClickHouseQuery.sqlTemplates() overrides boolean/timestamp but not string — so the base STRING leaks through.

Fix

Override templates.types.string = 'String' in ClickHouseQuery.sqlTemplates(), mirroring the existing boolean/timestamp overrides.

Note: #10316 also reported a null-safe IS NULL join-key bug, which was already fixed in v1.6.24 (#10494). This PR addresses the remaining string-cast-casing symptom.

[hughtopping]

Hi @igorlukanin - we've found that we get a different issue when String is added here. We're getting the following error from Clickhouse: Cannot convert NULL to a non-nullable type.

Using Nullable(String) rather than String fixes the error.

Here's a repro test case:

Clickhouse

CREATE TABLE buyer (id String, first_name String, last_name String, email_addresses Array(String)) ENGINE = MergeTree() ORDER BY id;
CREATE TABLE purchase (id String, buyer_id String) ENGINE = MergeTree() ORDER BY id;

INSERT INTO buyer VALUES ('1', 'Alice', 'Smith', ['alice@example.com', 'alice.smith@example.com']);
INSERT INTO purchase VALUES ('1', '1');

Cube

# purchases.yml
cubes:
  - name: purchases
    sql: SELECT id, buyer_id FROM purchase
    joins:
      - name: buyers
        sql: "{purchases}.buyer_id = {buyers}.id"
        relationship: many_to_one
    dimensions:
      - name: id
        sql: id
        type: string
        primary_key: true

# buyers.yml
cubes:
  - name: buyers
    sql: SELECT id, first_name, last_name FROM buyer
    joins:
      - name: purchases
        sql: "{buyers}.id = {purchases}.buyer_id"
        relationship: one_to_many
      - name: buyer_email_addresses
        sql: "{buyers}.id = {buyer_email_addresses}.buyer_id"
        relationship: one_to_many
    dimensions:
      - name: id
        sql: id
        type: string
        primary_key: true
      - name: first_name
        sql: first_name
        type: string

# buyer_email_addresses.yml
cubes:
  - name: buyer_email_addresses
    sql: |
      SELECT id AS buyer_id, arrayJoin(email_addresses) AS email_address
      FROM buyer
    joins:
      - name: buyers
        sql: "{buyer_email_addresses}.buyer_id = {buyers}.id"
        relationship: many_to_one
    dimensions:
      - name: id
        sql: buyer_id
        type: string
        primary_key: true
        public: false
      - name: email_address
        sql: email_address
        type: string
    measures: []

Cube SQL API query

SELECT
    purchases.id,
    buyers.first_name,
    array_agg(buyer_email_addresses.email_address) AS email_addresses
FROM purchases
CROSS JOIN buyers
CROSS JOIN buyer_email_addresses
WHERE purchases.id = '1'
GROUP BY purchases.id, buyers.first_name

Edit: this was also reported by another user in #10415 (comment)