fix(c-api): ZSTD_sizeof_CCtx underreports dictionary-backed contexts

[polaz]

Context

Found during review of #387.

Problem

ZSTD_sizeof_CCtx() sums only the inline ZSTD_CCtx struct and the scratch
buffer. After the first ZSTD_compress_usingCDict() call, the context also
caches a FrameCompressor (dict_compressor) whose primed match-finder tables
and dictionary snapshot dominate the real footprint, so the reported size can
miss most of the context's heap. Upstream ZSTD_sizeof_CCtx includes the
CDict-copied working tables.

Suggested approach

Add a heap-size accessor to the encoder that sums its owned allocations, then
include it in ZSTD_sizeof_CCtx when dict_compressor is Some:

• MatchGeneratorDriver::heap_size() — sum vec_pool capacities + the active
  MatcherStorage variant's owned buffers + the primed snapshot.
• A size method on each backend generator (FastKernelMatcher,
  DfastMatchGenerator, RowMatchGenerator, HcMatchGenerator) summing its
  hash/chain/history table capacities.
• FrameCompressor::heap_size() = matcher size + its own scratch/pending/FSE
  buffers + dictionary snapshot.

Mirrors the decoder-side FrameDecoder::workspace_size() already used by
ZSTD_sizeof_DCtx.

Files involved

• c-api/src/context.rs — ZSTD_sizeof_CCtx
• zstd/src/encoding/frame_compressor.rs — new heap_size
• zstd/src/encoding/match_generator.rs — driver heap_size
• zstd/src/encoding/simple/fast_matcher.rs, zstd/src/encoding/dfast/mod.rs,
  zstd/src/encoding/row/mod.rs — per-backend size methods

Why not in #387

A correct accounting requires new size methods across all four backend
generator modules (none touched by #387) plus snapshot recursion — a
self-contained feature, not a one-line fix. ZSTD_sizeof_CCtx is an estimate
API (no correctness/round-trip impact), so an incomplete hand-rolled sum would
be more misleading than tracking it properly here.

[coderabbitai]

🔗 Related PRs

#60 - feat(encoding): write frame content size in encoder output [merged]
#282 - perf(decode): lazy ring-buffer allocation for direct-eligible frames [merged]
#314 - perf(encode): skip decoder-table build when loading a dict for encoding [merged]
#328 - perf(encode): size-gated dictionary match-finding for the Fast strategy [merged]
#342 - perf(encode): cut small-input fixed cost on the fast levels [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[polaz]

Fixed in place in #387 rather than deferred: ZSTD_sizeof_CCtx now counts the cached dict compressor's heap via a new Matcher::heap_size() accessor (match-finder tables, recycled pool, primed snapshot, dict tables, BT/LDM scratch) plus the retained dictionary content. Regression test asserts the reported size grows once the primed tables exist.