Fix single embed_sparse and embed_all gRPC request metrics

[shoemoney]

What does this PR do?

The gRPC embed_sparse and embed_all handlers increment te_request_count{method="single"} twice per request, once before inference and again after a successful response, and never record te_request_success{method="single"}. One successful request therefore counts as two, and a success/error ratio derived from these counters reports a permanent 100% failure rate for sparse and all-embedding gRPC traffic.

This PR records the post-success event as te_request_success instead, at router/src/grpc/server.rs:647 and router/src/grpc/server.rs:689 on main. The sibling paths already implement the count then success pair correctly: the same file's embed method (grpc/server.rs 595/606) and the HTTP embed handlers (http/server.rs 603/621, 792/808, 970/986, 1174/1192). #917 fixes the identical double-increment in the HTTP /predict single arm and its description explicitly leaves the gRPC paths out of scope; this PR covers that named remainder.

Following the pattern of #917, the two counter calls move into record_single_request/record_single_success helpers with a recorder-backed unit test that needs no model, network, or accelerator. With the buggy counter name in place the test fails with count=2 and success=0; with this fix it passes. The already-correct embed, predict, rerank, and batch call sites are intentionally unchanged to keep the diff minimal.

Related to #879.

Validation

• cargo test -p text-embeddings-router --no-default-features --features grpc,candle --lib (1 passed)
• rustfmt --check router/src/grpc/server.rs --edition 2021
• cargo clippy -p text-embeddings-router --no-default-features --features grpc,candle --lib (passes; only pre-existing warnings, including result_large_err on the untouched auth closure at grpc/server.rs:1484)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Was this discussed/approved via a GitHub issue or the forum? Tracked in te_request_count (method=single) is incremented twice for single requests #879.
• Did you make sure to update the documentation with your changes? No documentation change is needed for this metrics correction.
• Did you write any new necessary tests? If applicable, did you include or update the insta snapshots? Added a recorder-backed unit test; no snapshot changes needed.