[iuliiasobolevska]

JavaDoc escaping introduced in #984 for #772 scans every description and repeats work for field, getter and builder documentation. Return ordinary descriptions unchanged, restrict markup parsing and Unicode regex scans to relevant input and reuse escaped descriptions within each generator instance. The HTML, inline-literal, comment-terminator and Unicode escaping rules remain unchanged.

Local comparison against 88728bd6, using the same large-schema fixture and byte-identical third-party dependency jars:

Workload	Before median	After median	Change
Construct, parse and generate without writes	525 ms	400 ms	-23.8%
Construct, parse, generate and write files	1,118 ms	1,039 ms	-7.1%

JDK 17, G1, a 2 GiB maximum heap and eight writers; two fresh JVMs per variant with six warmups and ten measured iterations each, in before/after/after/before order.
The improvement repeated in both fork comparisons, although write-inclusive timing ranges overlap.
Main-thread allocation fell by 41 MB without writes and 56 MB with writes.
All 9,952 generated source and example files (41,945,234 bytes) are byte-identical and share the same path-and-content SHA-256 digest.