perf: Lazily load fast fields dictionaries. (#2842)

## What

Lazily load fast field dictionaries from buffers: see
paradedb/tantivy#55

## Why

A customer reported slower-than-expected paging on a string/uuid column.
85% of the time for that query was being spent in _opening_ a fast
fields string/bytes column, with a large fraction of that time spent
fully consuming the column's `Dictionary`.

## Tests

See the attached benchmark results:
* [`docs`
dataset](#2842 (review))
    * No regressions.
    * 2x faster for `top_n-score`
    * 1.4x faster for `highlighting` 
* [`logs`
dataset](#2842 (review))
    * No regressions.
    * 4.5x faster for `paging-string-max`
    * 1.7x faster for `paging-string-median`
    * 1.6x faster for `paging-string-min`

The `paging-string-*` benchmarks were added in #2834 to highlight this
particular issue.

Author: stuhood

Cargo.lock

@@ -23,7 +23,7 @@ lto = "thin"
 codegen-units = 32
 
 [workspace.dependencies]
-tantivy = { git = "https://github.com/paradedb/tantivy.git", package = "tantivy", rev = "d37e61e6b8ddb105cbdc3ab391633ca108de6b73", features = [
+tantivy = { git = "https://github.com/paradedb/tantivy.git", package = "tantivy", rev = "16740add9447330b133348448ce4988edcc8174f", features = [
   "quickwit",        # for sstable support
   "stopwords",
   "lz4-compression",
@@ -34,4 +34,4 @@ tantivy-jieba = "0.11.0"
 
 [patch.crates-io]
 rust_icu_sys = { git = "https://github.com/google/rust_icu.git", rev = "53e98c8" }
-tantivy-tokenizer-api = { git = "https://github.com/paradedb/tantivy.git", package = "tantivy-tokenizer-api", rev = "d37e61e6b8ddb105cbdc3ab391633ca108de6b73" }
+tantivy-tokenizer-api = { git = "https://github.com/paradedb/tantivy.git", package = "tantivy-tokenizer-api", rev = "16740add9447330b133348448ce4988edcc8174f" }