panic: floats: slice lengths do not match

**Gorse version**
**0.5.0-alpha**, but also happens in **0.4.14**

**Describe the bug**
Every one hour or so we're encountering the following panic:
```go
panic: floats: slice lengths do not match

goroutine 104 [running]:
github.com/zhenghaoz/gorse/base/floats.Dot(...)
/go/gorse/base/floats/floats.go:169
github.com/zhenghaoz/gorse/base/search.(*DenseVector).Distance(0x21f5e80?, {0x21f4b38?, 0xc06508c140?})
/go/gorse/base/search/index.go:58 +0x2d4
github.com/zhenghaoz/gorse/base/search.(*HNSW).distance(0xc10dc3a0b0, {0x21f4b38, 0xc06508c140}, {0xc04a8d9574, 0x1, 0xc000509000?})
/go/gorse/base/search/hnsw.go:302 +0xc2
github.com/zhenghaoz/gorse/base/search.(*HNSW).efSearch(0xc10dc3a0b0, {0x21f4b38, 0xc06508c140}, 0xa?)
/go/gorse/base/search/hnsw.go:443 +0x4d
github.com/zhenghaoz/gorse/base/search.(*HNSW).MultiSearch(0xc10dc3a0b0, {0x21f4b38, 0xc06508c140}, {0xc0217ad880, 0x1b, 0x4a8d96d8?}, 0x105, 0x0)
/go/gorse/base/search/hnsw.go:421 +0x1fe
github.com/zhenghaoz/gorse/worker.(*Worker).collaborativeRecommendHNSW(0xc00056ca80, 0xc0216404d0?, {0xc018fd8ed8, 0x8}, {0xc0217ad880, 0x1b, 0x1b}, {0x220c4a8, 0xc055206de0}, 0xc01d81e8c0)
/go/gorse/worker/worker.go:878 +0x1e5
github.com/zhenghaoz/gorse/worker.(*Worker).Recommend.func2(0x0, 0x4e1)
/go/gorse/worker/worker.go:597 +0xb91
github.com/zhenghaoz/gorse/base/parallel.Parallel(0x48892, 0xc021c38de0?, 0xc04a89c280)
/go/gorse/base/parallel/parallel.go:40 +0xf7
github.com/zhenghaoz/gorse/worker.(*Worker).Recommend(0xc00056ca80, {0xc0584ec000, 0x48892, 0x4b238})
/go/gorse/worker/worker.go:550 +0xccf
github.com/zhenghaoz/gorse/worker.(*Worker).Serve.func1()
/go/gorse/worker/worker.go:425 +0x405
github.com/zhenghaoz/gorse/worker.(*Worker).Serve(0xc00056ca80)
/go/gorse/worker/worker.go:437 +0xddf
main.glob..func1.1()
/go/gorse/cmd/gorse-in-one/main.go:109 +0x3f
created by main.glob..func1 in goroutine 1
/go/gorse/cmd/gorse-in-one/main.go:107 +0xb1d
```

**To Reproduce**
I'm unfortunately not able to reproduce the bug reliably, it occurs randomly inside `woker.go`:
```go
if time.Since(tick) < w.Config.Recommend.Offline.CheckRecommendPeriod {
loop()
}
```

**Additional context**
- gorse-in-one Docker image
- MongoDB for cache and MySQL for item storage
- 360K items, 700K feedback, ~30 categories

**Config**

```toml
[master]
n_jobs = 16
meta_timeout = "10s"

[server]
default_n = 10
clock_error = "5s"
auto_insert_user = true
auto_insert_item = true
cache_expire = "10s"

[recommend]
cache_size = 250
cache_expire = "12h"
active_user_ttl = 1

[recommend.data_source]
positive_feedback_types = ["comment","like","favorite"]
read_feedback_types = ["read"]
positive_feedback_ttl = 7
item_ttl = 30

[recommend.popular]
popular_window = "24h"

[recommend.user_neighbors]
neighbor_type = "related"
enable_index = true
index_recall = 0.8
index_fit_epoch = 3

[recommend.item_neighbors]
neighbor_type = "related"
enable_index = true
index_recall = 0.8
index_fit_epoch = 3

[recommend.collaborative]
enable_index = true
index_recall = 0.9
index_fit_epoch = 3
model_fit_period = "60m"
model_search_period = "360m"
model_search_epoch = 100
model_search_trials = 10
enable_model_size_search = true

[recommend.replacement]
enable_replacement = false
positive_replacement_decay = 0.8
read_replacement_decay = 0.6

[recommend.offline]
check_recommend_period = "15m"
refresh_recommend_period = "24h"
enable_latest_recommend = true
enable_popular_recommend = true
enable_user_based_recommend = true
enable_item_based_recommend = true
enable_collaborative_recommend = true
enable_click_through_prediction = true
explore_recommend = { popular = 0.2, latest = 0.1 }

[recommend.online]
fallback_recommend = ["item_based", "popular"]
num_feedback_fallback_item_based = 10

[tracing]
enable_tracing = false
exporter = "jaeger"
collector_endpoint = "http://localhost:14268/api/traces"
sampler = "always"
ratio = 1
```